Multimodal agents increasingly operate computers on behalf of users. This shift from passive question answering to interactive execution creates a new safety problem. When an agent encounters normal, non-attack scenarios where the task it needs to perform is impossible or difficult to complete, the central risk is not only that it fails, but that it deceives the user into believing that it has succeeded. Although prior work has studied truthfulness and deception in language models and language agents, multimodal interaction introduces new sources of deceptive behavior, increasing the implicitness and complexity of deceptive behaviors in multimodal agents. To fill this gap, we introduce $\mathbf{\mathtt{MAD\text{-}Bench}}$, a benchmark for evaluating **M**ultimodal **A**gent **D**eception. $\mathbf{\mathtt{MAD\text{-}Bench}}$ contains 360 sandbox tasks organized around three core elements of multimodal execution: *modality*, *target*, and *tool*, and spans six task types including *modality evidence conflict*, *modality asynchronous mismatch*, *modality ambiguity distortion*, *modality object missing*, *target infeasibility*, and *tool defect*. To our knowledge, $\mathbf{\mathtt{MAD\text{-}Bench}}$ is the first benchmark to formalize deceptive behavior in multimodal agents from an execution-evidence perspective, characterizing deception as a misalignment between the user-facing task-state claim and the multimodal evidence available throughout the execution trajectory. We further propose a behavioral taxonomy covering *evasive deception*, *manipulative deception*, *misleading deception*, and *non-deception*. Evaluating 10 mainstream multimodal agents, we find that agents frequently fabricate, conceal, mislead, frame, and disguise during task execution. These results show that substantial deceptive behaviors are widespread across mainstream model families. This benchmark is available at https://anonymous.4open.science/r/MAD-Bench-A304.

Designing effective reward functions is a critical challenge in Reinforcement Learning (RL), which traditionally requires costly manual trial-and-error. While Large Language Model (LLM)-based methods have shown promise in reward design, undifferentiated sampling and limited use of historical feedback often lead to insufficient exploration of the reward space and optimization instability. To address these challenges, we propose $R^2E$, a Role-driven Reward Evolutionary framework for automated reward function design that structures the process through explicit role specialization. $R^2E$ employs multiple LLM roles with complementary objectives: an Explorer that promotes novelty to expand the search space, a Guardian that performs conservative refinements to improve training stability, and an Evolver that recombines high-performing rewards via crossover and mutation to integrate effective structures. A dynamic scheduling strategy coordinates these roles, progressively shifting the search from broad exploration to focused exploitation. To mitigate optimization instability and provide a medium for cross-role collaboration, $R^2E$ incorporates a Global Elite Pool that retains the best-performing rewards to guide subsequent generations, thereby enabling coordinated interaction among roles and ensuring a reliable and consistent refinement process. Extensive experiments on multiple robotic tasks demonstrate the effectiveness of our method.

Vehicle trajectory generation is critical for autonomous driving simulation. However, current generative models face two coupled limitations: (i) imbalanced trajectory distributions bias them toward frequent motions, such as straight driving or stationary states, while weakening their ability to synthesize rare but safety critical motions, such as turning. (ii) Euclidean intermediate spaces provide limited geometric inductive bias for modeling vehicle pose evolution. Together, these limitations lead to mode biased generation and physically less coherent trajectories. To address this issue, we propose SEA, an \underline{\textbf{SE}}(2)-\underline{\textbf{A}}ware vehicle trajectory generation method that realizes the conditional transport path through planar rigid body poses. At the \textbf{data distribution} level, SEA learns vehicle trajectory distributions from traffic data to capture individual motion characteristics. At the \textbf{physical-space} level, motivated by the geometry dependence of {\bf optimal transport} interpolation, SEA realizes intermediate transport states through an $SE(2)$-consistent construction, thereby preserving geometric interpretability and physical coherence during generation. Extensive experiments on large scale real world datasets show that our method outperforms state of the art approaches by generating more realistic, diverse and dynamically consistent trajectories.

As one of the most complex and load-bearing joints in the human body, the ankle plays a crucial role in locomotion and clinical assessment. Accurate segmentation of ankle bones is essential for trauma evaluation, preoperative planning, and disease diagnosis. However, the scarcity of publicly available ankle datasets has hindered the development of intelligent analysis in this area. To address this gap, we introduce the first two publicly available ankle CT datasets, providing new directions for medical image segmentation. Considering the intricate and spatially correlated anatomy of the ankle, we propose a 3D medical image segmentation model, namely 3DABSeg. Specifically, we introduce the Hybrid KAN-Mamba Block (HKMB) to capture long-range spatial dependencies and complex nonlinearities. By integrating Mamba’s sequence modeling with KAN’s learnable spline functions, HKMB enhances feature expressiveness for complex anatomical structures. Furthermore, to address the uneven semantic distribution and feature dilution across channels, we propose a Multi-Scale Feature Fusion Mixture-of-Experts module (MFF-MoE). MFF-MoE utilizes multi-scale spatial pooling to compute dynamic routing scores, adaptively partitioning channels into specialized expert networks via a mutually exclusive hard-assignment strategy. Extensive experiments demonstrate that 3DABSeg surpasses existing state-of-the-art methods in both segmentation performance and boundary precision, highlighting its potential for complex anatomical structure analysis and clinical applications.

Current embodied VLM evaluation relies on static, expert-defined, manually annotated benchmarks that exhibit severe redundancy and coverage imbalance. This labor‑intensive paradigm drains computational and annotation resources, inflates costs, and distorts model rankings, ultimately stifling iterative development. To address this, we propose Agentic Automatic Evaluation (A2Eval), the first agentic framework that automates benchmark curation and evaluation through two collaborative agents. The Data Agent autonomously induces capability dimensions and assembles a balanced, compact evaluation suite, while the Eval Agent synthesizes and validates executable evaluation pipelines, enabling fully autonomous, high-fidelity assessment. Evaluated across 10 benchmarks and 32 models, A2Eval compresses evaluation suites by 85\%, reduces overall computational costs by 77\%, and delivers up to a 4.6$\times$ speedup while preserving evaluation quality. Crucially, A2Eval corrects systematic ranking biases, improves human alignment to Spearman's $\rho=0.85$, and maintains high ranking fidelity (Spearman's $\rho=0.87$), establishing a new standard for high-fidelity, low-cost embodied assessment. Our code and data will be public upon acceptance.

Optimal transport (OT) provides a principled framework for mapping between probability distributions. Despite extensive progress, applying OT to large-scale data remains computationally demanding, and the resulting pointwise transport plans are often difficult to interpret. We introduce Optimal Mixture Transport (OMT), a scalable framework that shifts the transport paradigm from individual samples to mixtures of subpopulations, reformulating the transport problem as a strictly biconvex optimization with a unique global minimizer. We further establish theoretical guarantees on the stability of the OMT map, showing that bounded perturbations of the underlying distributions lead to bounded changes in the transport plan. By formulating subpopulations as exponential-family distributions, OMT decouples computational complexity from the sample size, scaling solely with the number of mixture components. We demonstrate the effectiveness and practicality of OMT on a wide range of synthetic benchmarks and real-world datasets, including image data and large-scale single-cell RNA sequencing measurements.

Online Conformal Prediction (OCP) sequentially constructs prediction sets whose average empirical coverage converges to a target level over time, for arbitrary, potentially adversarial datasets. In the multi-level setting of OCP, prediction sets are produced at $d$ coverage levels at once. Prior work showed that online gradient descent with lazy isotonic projections produces sets properly nested across levels, but with a $\sqrt{d}$ dependence in the per-level coverage convergence rate. Here, we both generalize and improve prior work. First, we show that lazy isotonic projections can be safely used for the construction of multi-level online conformal prediction sets from \emph{any} online learning algorithm. Then, through a black-box reduction, we show that the linearized regret of the online algorithm itself automatically gives control over both per-level calibration and distributional consistency of multi-level predictions via Fenchel duality. Finally, by using the Weighted Interval Score, the standard scoring rule for assessing multi-level interval forecasts, we prove that these projections reduce both the size of the sets and their under-coverage. We show two instantiations of our theory, $p$-norm Online Mirror Descent and coordinate-wise Universal-Portfolio, both improving the dependence on the number of levels to $\sqrt{\log d}$ for the multi-OCP problem. Experiments on stock prices, electricity demand, and synthetic streams confirm consistent gains in coverage-width trade-offs over existing baselines.

Pre-training Large Language Models requires immense computational resources, making optimizer efficiency essential. The optimization landscape is highly anisotropic, with loss reduction driven predominantly by progress along flat directions. While matrix-based optimizers such as Muon and SOAP use fine-grained curvature information to outperform AdamW, their updates tend toward isotropy, which is relatively conservative along flat directions yet potentially aggressive along sharp ones. To address this limitation, we first establish a unified Riemannian Ordinary Differential Equation (ODE) framework that clarifies the mechanism of adaptive algorithms: the preconditioner induces a Riemannian geometry that mitigates ill-conditioning, while momentum serves as a Riemannian damping term that promotes convergence. Guided by these insights, we propose LITE, a generalized acceleration strategy that enhances training dynamics by applying larger Hessian damping coefficients and learning rates along flat trajectories. Extensive experiments demonstrate that LITE significantly accelerates both Muon and SOAP across diverse architectures (Dense, MoE), parameter scales (130M to 1.3B), datasets (C4, Pile), and learning-rate schedules (cosine, wsd). Theoretical analysis confirms that LITE facilitates faster convergence along flat directions in anisotropic landscapes, providing a principled approach to efficient LLM pre-training.

Deep learning inference demand is projected to grow $10{,}000\times$ over the next five years, a trajectory that general-purpose accelerators cannot match. Custom accelerators offer a viable path forward, but designing them requires simultaneous co-optimization of hardware datapaths, software schedules, and compiler transforms across a combinatorial search space exceeding $\mathcal O(10^{2300})$. While co-design is becoming increasingly essential, existing methodologies rely on decoupled, sequential pipelines that miss critical cross-stage trade-offs. We introduce CHEETAH: a globally optimized, AI-driven framework for inference accelerator co-design. To our knowledge, this is the first co-design framework to combine \textit{joint} tiling-fusion, dynamic placement, and rematerialization in a single LP-based formulation. Key contributions include: (1) time-indexed variables for dynamic memory management, (2) rematerialization variables to enable cheap recomputation over costly DRAM reloads, and (3) joint tiling-fusion optimization via McCormick linearization. This inner-loop precision is steered by an LLM-guided outer loop that performs causal interventions based on specific deployment profiles and Lagrangian sensitivity signals. In less than 6 hours, \textsc{CHEETAH} discovers designs on the Pareto frontier achieving an average of $\sim22.07\times$ raw speedup across $11$ workloads, reaching up to $\sim40.1\times$ speedup on a single workload over a TPU-v3 baseline, unlocking efficiency inaccessible to decoupled sequential pipelines.

3D Gaussian Splatting (3DGS) has emerged as a powerful paradigm for real-time photorealistic novel view synthesis. However, its training-time densification remains largely heuristic, often resulting in redundant Gaussian proliferation and inaccurate geometric reconstruction. Existing methods predominantly rely on first-order screen-space gradients to trigger cloning and splitting, which fail to distinguish true geometric discontinuities from high-frequency textures or noisy residuals. To address this limitation, we propose ACD-GS, an Asymmetric Curvature-aware Densification framework for compact and geometry-aware 3D Gaussian optimization. Our key insight is that second-order depth responses provide a more reliable structural prior for density control, enabling asymmetric regulation of cloning and splitting to better capture geometric discontinuities. Building on this, we introduce a multi-view photometric consistency constraint to suppress transient single-view errors, together with a capacity-aware adaptive regulation strategy that balances densification and pruning during training. Unlike post-hoc compression or quantization-based approaches, ACD-GS directly prevents redundant Gaussian generation during training, yielding inherently compact representations. Extensive experiments demonstrate that our method reduces nearly 60\% of Gaussian count and storage while maintaining competitive rendering fidelity, consistently outperforming existing methods across multiple datasets. Our code is available in https://anonymous.4open.science/r/ACD_GS-0E2C

Causal reasoning from observational data becomes significantly challenging when the underlying causal structure is only partially known. In this work, we study the the equivalence class of structural causal models (SCMs) with latent variables sharing the same set of conditional independencies, as represented by a partial ancestral graph (PAG). Specifically, we provide a new characterization of SCM-induced causal diagrams in this equivalence class using differentiable parameters. Building on this characterization, we introduce a new model class called PAG-constrained neural causal models ($\mathcal{P}$-NCMs) which parameterize this equivalence class of SCMs. We prove that $\mathcal{P}$-NCMs are expressive enough to represent counterfactual distributions induced by any SCM in the class, while remaining consistent with the structural constraints shared across all its members. Finally, we demonstrate how the differentiable parameterization of this model class enables causal inference under Markov equivalence by reducing counterfactual partial identification to an optimization problem over $\mathcal{P}$-NCMs. We establish the theoretical soundness of this approach and validate its performance on simulations.

Large language models face substantial computational challenges in long-sequence inference due to the quadratic cost of self-attention, with the prefill stage being particularly bottlenecked. Static sparse attention reduces compute via predefined masks but can miss task-critical evidence, while dynamic sparsification is more flexible yet can be brittle across tasks, leading to either insufficient coverage or distractor-induced noise. We propose AdaCal, an Adaptive Calibration framework for robust sparse attention that synergizes coarse, experience-driven task dispatch with head-wise physical profiling. By deriving entropy, concentration, and mid-zone coverage metrics from a lightweight probe pattern, AdaCal dynamically governs head activation and augmentation policies, allocating additional computation strictly on demand. This design is realized through task-conditional policies that effectively suppress unnecessary expansion for structure-dominant inputs while enhancing long-range coverage for retrieval-intensive tasks. Experiments show that AdaCal improves robustness over strong static and dynamic baselines, with clearer gains in long-context settings beyond 32K tokens.

It has been widely observed that Adam can remain stable even when the objective function deviates significantly from global smoothness. However, under the generalized smoothness framework, existing theoretical analyses typically rely on strong tail assumptions on stochastic gradients, such as almost-sure boundedness or sub-Gaussianity. Whether one can establish the convergence of Adam on generalized smooth objectives under only second moment information on the stochastic gradients, without imposing such strong concentration assumptions, was explicitly identified as an important open direction by \citet{li2023convex}. This paper gives an affirmative answer to this question under fairly general conditions. Specifically, we prove that such strong tail assumptions are not necessary. We show that the key mechanism by which Adam remains stable and achieves convergence under the (L0)--(Lp) generalized smoothness condition is the self-normalization effect induced by its adaptive coordinate-wise scaling. Based on this mechanism, we prove that even under a very general stochastic-gradient condition, namely a generalized second moment ABC condition that provides only second moment information, the stochastic trajectory of Adam remains in a locally well-behaved smoothness region with stretched-exponential tail decay. As a consequence, we establish high-probability convergence rate guarantees over the full range (p<2), with a confidence dependence of order (\delta^{-1/2}), while the required stepsize calibration depends on (\delta) only through polylogarithmic factors in (\polylog(1/\delta)). Furthermore, we construct a hard instance proving that, under only second-moment information on the stochastic gradients, this (\delta^{-1/2})-type confidence dependence is sharp. Finally, in the more favorable regime (p<1), we combine the above trajectory control with polynomial-growth estimates on rare events to further obtain convergence rate guarantees in expectation.

LM-based agents process external content, exposing them to prompt injection and multi-turn manipulation. Most safety benchmarks evaluate defenders against fixed attack pools collected before evaluation, single-turn or multi-turn. We present a 21-scenario benchmark for adaptive multi-round attacks against memoryless LLM defenders: an autonomous LLM attacker observes prior defender responses and pivots across rounds, while each defender response is evaluated as a fresh interaction. Holding the 21 scenarios, attackers, defenders, and structured-output scoring fixed, restricting scoring to the first attacker turn yields $0$--$1%$ attack success rate (ASR); allowing 15 rounds of adaptive attack yields $8.3$--$19.0%$. Pooling three frontier attacker LLMs uncovers up to $2.4\times$ as many unique successful attacks as the best single attacker, and the generated attacks have low cosine similarity ($0.02$--$0.14$) to attacks in existing benchmarks. Claude Opus 4.6 and GPT-5.4 are aggregate-equivalent ($8.3%$ vs. $11.1%$ ASR), but their weaknesses differ sharply: on one scenario Opus reaches $60%$ ASR ($95%$ CI $36$--$80%$) while GPT-5.4 and Gemini each stay at $7%$ (CI $1$--$30%$). $15$ of $21$ scenarios distinguish at least one defender pair, yet rankings disagree across scenarios (Kendall's $W = 0.21$). We release the benchmark---21 evaluation scenarios, 10 public development scenarios, the orchestrator, baseline harnesses, and a multi-attacker CLI---plus 945 transcripts from the $3\times3$ frontier matrix, an attack-replay dataset, and $18{,}422$ battles from an open competition's final scoring rounds.

Evolution strategies (ES) have recently emerged as a promising alternative to gradient-based reinforcement learning for fine-tuning large language models, but classical zeroth-order optimization theory does not systematically account for the empirical observations driving this success. We formulate the continuous-time limit of Z-score-normalized ES as a unified stochastic differential equation and identify three sequential dynamical regimes characterized by the dominant gradient contribution to the population reward variance; under a basin-local low-rank Hessian assumption, this framework systematically explains (1) the information-geometric optimality of the Z-score heuristic, (2) the dimensionality paradox of sample-efficient ES with populations far smaller than the parameter dimension, (3) the same-task rise-then-decay of the training reward, and (4) catastrophic forgetting of pre-training capabilities. Synthetic experiments and direct verification on Qwen2.5-{0.5B,3B,7B}-Instruct confirm these predictions, matching the parameter drift reported by Abdi et al. (2026) within $\sim 1.5\\%$.

Existing audio deepfake detection (ADD) methods frequently struggle with limited generalization, as fixed bottleneck architectures fail to adapt to the heterogeneous information densities inherent in diverse spoofing artifacts. Moreover, standard augmentation often incurs a pathological robustness-accuracy trade-off. In this paper, we propose a unified ADD framework to address these challenges. Our approach integrates: (1) an Adaptive Audio Feature Integration (AAFI) module that dynamically selects optimal latent dimensions via a compression pool and a stochastic exploration mechanism, ensuring flexible representations across multi-type dynamic attacks; and (2) a Mel-Band Adversarial Perturbation (Mel-BAP) strategy that applies targeted regularization in the mel-spectrogram domain to encourage the learning of invariant decision boundaries. Extensive evaluations on benchmarks including ADD 2022, FakeOrReal, and In-the-Wild demonstrate that our Whisper-small based model achieves a state-of-the-art average Equal Error Rate (EER) of 1.51\%. Compared to existing SOTA models such as the 3-billion-parameter Resemble-Detect-3B-Omni, our framework reduces parameter count by up to 91.6\% while improving average EER by 60.7\%. Notably, the model exhibits exceptional zero-shot transferability, achieving 3.91--5.65\% EER on unseen languages and maintaining high resilience against real-world acoustic distortions. This work provides a scalable and efficient “plug-in” solution for enhancing trust and security in the audio Web ecosystem.

We study convex optimization problems over a compact convex set where projections are expensive but a linear minimization oracle (LMO) is available. We propose the *adaptive conditional gradient sliding method* (AdCGS), a projection-free and line-search-free method that retains Nesterov's acceleration with adaptive stepsizes based on local Lipschitz estimates. AdCGS combines an accelerated outer scheme with an LMO-based inner routine. It reuses gradients across multiple LMO calls to reduce gradient evaluations, while controlling the subproblem inexactness via a prescribed accuracy level coupled with adaptive stepsizes. We prove accelerated rates for convex objective functions, matching projection-based methods, without relying on a projection oracle. For locally strongly convex objective functions, we further establish linear convergence without additional geometric assumptions on the constraint set, such as polytopes or strongly convex sets. Experiments on constrained $\ell_p$ regression, logistic regression, and least-squares problems demonstrate that AdCGS improves over projection-free baselines and provides competitive performance when projections are inexpensive.

Multi-turn conversation is the predominant form of interaction with large language models (LLMs), where user queries continuously evolve. However, existing LLM routing methods are primarily designed for single-turn interactions or fixed queries settings, overlooking the dynamic nature of multi-turn conversations and the challenge of delayed rewards, thereby limiting their ability to optimize cumulative performance. To address this challenge, we move from myopic, single-turn selection to long-horizon routing for multi-turn conversation. Accordingly, we propose ConvoRouter, which first performs MCTS to explore conversation branches induced by different LLM selections and collect trajectories with high cumulative rewards. ConvoRouter then learns a lightweight routing policy from search-derived data, augmented with retrieval-based future state approximation, enabling multi-turn routing without online search. Experiments on both open-domain and domain-specific conversation tasks across diverse candidate sets of both open-source and closed-source LLMs demonstrate that ConvoRouter significantly outperforms single LLMs and existing routing baselines in task success rate, while achieving a superior performance-cost trade-off when combined with a cost-aware reward.

Real-world tasks rely on information from multiple sensory sources, motivating multimodal learning as a core paradigm in modern machine learning. However, multimodal learning suffers from modality laziness, where one modality suppresses the contribution of the other modalities by dominating the training process. Recent studies of proposing various metrics to identify lazy modalities and manipulate optimizations to mitigate this imbalance, facing two main problems: 1) The identification of lazy modality is incomplete and overly sharp; 2) The dominated modality can be lagged by under-optimized lazy modalities. To address these issues, we propose Adaptive-Margin Masking and Restoration (AMRe), which introduces adaptive-margin modality identification and restoration optimization to balance dominated and lazy modalities. Experimental results show that AMRe consistently outperforms competitive baselines and several state-of-the-art methods by achieving significant improvement on standard multimodal benchmarks. The codes are released in https://anonymous.4open.science/r/AMRe-5E69.

Modern deep reinforcement learning algorithms often store past sensory observations in a buffer for later replay and learning. However, this process is biologically implausible; animals do not store raw sensory observations. Instead, they store their imperfect understanding of the world and the events in it. An alternative to storing all past sensory observations is to learn directly from the observations as they come in and store the seemingly important parts of the incoming data. Eligibility trace algorithms in reinforcement learning (RL) sweep back updates for multiple steps, improving efficiency and providing update rules that do not rely on storing large data buffers. These algorithms were common and effectively used in the linear setting, but have not been widely adopted in the deep RL setting. Many questions remain open to make these algorithms more effective, even as basic as how to incorporate vector step-sizes. We first highlight how adaptive vector step-sizes need to be incorporated into the eligibility trace and then introduce a new vector step-size algorithm that is more amenable to the forward-backward view equivalence needed to derive an update with eligibility traces. We show that our new algorithms improve over other streaming algorithms across Mujoco and MinAtar environments, and are comparable to methods with replay buffers such as PPO.

Primal-dual interior-point methods solve constrained convex optimization problems to tight tolerances with speed and robustness. Their solutions are also efficiently differentiable with respect to the problem data through the implicit function theorem. However, the standard treatment of primal-dual complementarity makes the underlying linear systems increasingly ill-conditioned near the solution. While this ill-conditioning is often benign in double precision, it can be catastrophic in single precision, preventing interior-point methods from fully exploiting the accelerated hardware that underpins modern machine learning. This paper introduces a differentiable interior-point method designed for low-precision arithmetic. By using an alternative complementarity representation, we ensure that the underlying linear systems remain spectrally bounded --- even near the solution --- a property that is essential for computing accurate gradients and avoiding arithmetic exceptions. As a result, our method enables interior-point solvers to reliably solve and differentiate optimization problems in single precision that were previously confined to double precision. We demonstrate the approach through an ablation study against the standard interior-point formulation and applications in bilevel and end-to-end learning settings where differentiating through constrained optimization is essential.

Autoregressive video generation has improved rapidly in visual fidelity and interactivity, but it still suffers from long-term inconsistency and memory degradation. Most existing solutions either compress historical frames using predefined strategies or retrieve keyframes based on coarse implicit attention signals, both of which fail to handle evolving prompts with shifting entity references, leading to identity drift, character duplication, and attribute loss. To address this, we propose IAMFlow, a training-free identity-aware memory framework that explicitly models and tracks persistent entity identities, enabling consistent generation across prompt transitions. Specifically, an LLM extracts entities with visual attributes from each prompt and assigns unique global IDs for identity-aware memory, while a VLM asynchronously verifies and refines attributes from rendered frames, enabling explicit entity tracking in place of implicit similarity-based matching. To keep the proposed framework computationally practical, we design a systematic inference acceleration pipeline, including asynchronous visual verification, adaptive prompt transition, and model quantization, which achieves faster generation than existing baselines. Furthermore, we introduce NarraStream-Bench, a benchmark for narrative streaming video generation that features 324 multi-prompt scripts spanning six dimensions and a three-dimensional evaluation protocol that integrates both traditional metrics and multimodal large language model-based assessments. Extensive experiments show that IAMFlow, despite being training-free, achieves the best overall performance on NarraStream-Bench, outperforming the strongest baseline by 2.56 points, while achieving a 1.39$\times$ speedup over the most efficient baseline in the 60-second multi-prompt setting.

Neural subgraph retrieval systems, which retrieve corpus graphs containing a query graph as a subgraph, are used in safety-critical applications such as drug discovery, hardware trojan detection, and molecular similarity search. Despite their importance, the adversarial robustness of these systems remains unstudied. We propose GRAP, the first adversarial attack framework targeting subgraph matching-based graph retrieval. \our jointly selects a budget-constrained subset of corpus graphs and computes per-graph edge perturbations to maximally degrade retrieval quality. We formalize two attack regimes: a ranking attack (with solver access) that corrupts pairwise relevance ordering while preserving relevance labels, and a solver-free top-K attack that directly displaces relevant results from retrieved sets. In both cases, we show that the resulting adversarial set functions are monotone and approximately submodular, enabling greedy subset selection with provable approximation guarantees. Experiments on five benchmark datasets against multiple state-of-the-art victim retrievers demonstrate that our method consistently and significantly outperforms all baselines in both gray-box and black-box settings, with and without solver access.

Adversarial Machine Learning research is currently centered on the search for the $\epsilon$-ball: minimal perturbations designed to fool models while remaining imperceptible to humans. We argue that this fixation on $\ell_p$ norms is a metrological trap that communicates a distorted perception of risk to defenders, one that is unaligned with real-world system vulnerabilities. This paradigm misleads defenders into optimizing for narrow, mathematical robustness at the expense of systemic security, while simultaneously incentivizing attackers to exploit non-overlapping alternative adversarial pathways. Ultimately, we argue that the dominance of the $\ell_p$ ball does not just fail to reduce harm - it may actively facilitate it by communicating a false sense of security against the wrong threats.

Current world models for explicit world simulation typically follow two distinct pathways: (1) video generative models (e.g., Sora), which simulate realistic appearances but fail to preserve 3D consistency; and (2) 3D generative models (e.g., Marble), which generate geometrically consistent scenes but often yield static, less realistic observations. We present AEON, a unified framework that bridges video-based and 3D-based world models and combines the benefits of both. The key insight of AEON lies in a generative model that operates within the video latent space while producing renderable 4D graphics primitives—specifically, 4D Gaussian Splatting (4DGS)—as the world modeler. We train a 4D reconstruction model to predict both 4DGS and camera poses from monocular videos, serving as our initialized 4D decoder. Furthermore, we design a domain adapter that translates generative video latents into reconstructive 4D latents through a distillation process on the video VAE. AEON naturally supports video-based world modeling by rendering 4DGS at predicted camera trajectories, and constructs a persistent 4D environment with scene dynamics for realistic world simulation. Consequently, AEON empowers a diverse range of world modeling applications and achieves state-of-the-art performance on standard benchmarks.

Accurate long-term air quality forecasting is challenging due to long-range transport, meteorological variability, and nonlinear chemical reactions underlying secondary pollutant formation. Purely data-driven models often suffer from long-horizon drift and lack physical consistency, while existing physics-informed approaches commonly follow open-loop designs that fuse neural and physical components only at the output level. We propose AeroChem, a closed-loop physics-informed state space framework for chemically reactive air quality forecasting. AeroChem adopts a two-branch co-correction design: a UDE-based process branch provides a structured but imperfect physical prior, while a Mamba-based virtual measurement branch captures long-range temporal context and produces probabilistic virtual observations. Neural Kalman Fusion couples the two branches by correcting the latent physical state at each forecasting step and feeding the posterior state back into the simulator, reducing long-horizon drift. We further introduce HanoiAir, a real-world dataset integrating pollutant concentrations, meteorology, and emission inventories. Experiments on HanoiAir and a public benchmark show that AeroChem improves long-horizon stability and sudden-change forecasting.

While bound propagation methods are highly effective for certifying neural network robustness against $\ell_{\infty}$ adversaries, scalable verification for $\ell_{2}$ perturbations remains a significant open challenge. Existing approaches suffer from severe geometric loss: elementwise bounding methods like $\alpha$-CROWN relax Euclidean balls to bounding boxes, losing a factor of $\sqrt{n}$ in radius, while the recently proposed SDP-CROWN rounds deformed ellipsoids back into balls, losing a factor proportional to the condition number of the weights. Consequently, tight $\ell_{2}$ verification is currently restricted to near-isometric or Lipschitz-regularized networks where these approximations are not vacuous. In this paper, we introduce affine-image propagation, a novel CROWN-compatible framework that exactly propagates the affine geometry of $\ell_{2}$ balls and $\ell_{\infty}$ boxes through the network. To bypass the prohibitive computational cost of exact Semidefinite Programming (SDP) constraints required by this representation, we derive highly scalable, SDP-based bounds utilizing diagonal-dominance surrogates. Our approach effectively eliminates the geometric bottlenecks of prior work, yielding the first large-scale, SDP-style bound propagation method capable of tight $\ell_{2}$ verification for general, non-Lipschitz-regularized neural networks.

AI agent research spans a wide spectrum: from RL agents that learn from scratch to foundation model agents that leverage pre-trained knowledge, yet no unified benchmark enables fair comparison across these approaches. We present Agentick, a benchmark for sequential decision-making agents designed to evaluate RL, LLM, VLM, hybrid, and human agents on common ground and to power research on the fundamental challenges of sequential decision-making. Agentick provides 37 procedurally generated tasks across six capability categories, four difficulty levels, and five observation modalities, all exposed through a single Gymnasium-compatible interface. The benchmark ships with a Coding API, oracle reference policies for all tasks, pre-built SFT datasets, a composable agent harness, and a live leaderboard. An evaluation spanning 27 configurations and over 90,000 episodes reveals that no single approach dominates: GPT-5 mini leads overall at 0.309 oracle-normalized score while PPO dominates planning and multi-agent tasks; the reasoning harness multiplies LLM performance by 3--10x; and ASCII observations consistently outperform natural language. These findings highlight the substantial room for improvement that remains across all agent paradigms. Agentick's capability-decomposed, multi-modal design provides the empirical infrastructure needed to drive progress toward general autonomous agents, both as an evaluation framework and as a training ground for RL post-training of foundation models in truly sequential environments.

The expansion of retrieval-augmented generation (RAG) into multimodal domains has intensified the challenge for processing complex visual documents, such as financial reports. While page-level chunking and retrieval is a natural starting point, it creates a critical bottleneck: delivering entire pages to the generator introduces excessive extraneous context. This not only overloads the generator's attention mechanism but also dilutes the most salient evidence. Moreover, compressing these information-rich pages into a limited visual token budget further increases the risk of hallucinations. To address this, we introduce AgenticOCR, a dynamic parsing approach that extends optical character recognition (OCR) from a static, full-text process into a query-driven, on-demand extraction system. By autonomously analyzing document layout in a "thinking with images" manner, AgenticOCR identifies and selectively recognizes regions of interest. This approach performs on-demand decompression of visual tokens precisely where needed, effectively decoupling retrieval granularity from rigid page-level chunking. AgenticOCR has the potential to serve as the "third building block" of the visual document RAG stack, operating alongside and enhancing standard Embedding and Reranking modules. Experimental results demonstrate that AgenticOCR improves both the accuracy and efficiency of visual RAG systems with pixel-adaptive per-image token allocation (e.g., Qwen3.5 generator), achieving expert-level performance in long document understanding. Our repository is anonymously available at https://anonymous.4open.science/r/AgenticOCR.

Reinforcement learning with verifiable rewards (RLVR) provides a promising pathway for continuously advancing GUI agents, yet existing reward modeling paradigms face complementary limitations. Rule-based methods suffer from poor scalability and cannot handle open-ended tasks. LLM-as-a-Judge methods enable scalable trajectory verification but remain passive and are constrained by partial state observability, since key evidence often resides in latent environment states beyond the trajectory. Recent active environment interaction methods mitigate observability issues but tend to over-rely on probing while under-utilizing direct trajectory evidence, leading to verification inefficiency. To address these challenges, we advocate a trajectory-grounded interactive verification paradigm. We introduce VAGEN, a framework that employs a tool-augmented verifier agent governed by a Progressive Verification Mechanism, which follows a surface-to-latent and cheap-to-expensive design philosophy to extract trajectory evidence and probe environment states in a proactive end-to-end manner. Experiments on OSWorld-Verified and AndroidWorld benchmarks demonstrate that VAGEN significantly improves evaluation accuracy with a favorable performance-efficiency trade-off.

Video production workflows offer a rich and demanding testbed for evaluating multimodal AI agents: they require composite capabilities across text, image, audio, and video understanding, along with long-horizon planning, and tool use. To this end, we introduce AgenticVBench, a benchmark of 100 tasks across 4 task families spanning the real world post-production workflow, constructed from the representative work of 20 industry experts averaging 6 years of experience across video production contexts. Tasks are paired with evaluation specifications that combine programmatic verifiers and expert rubrics, both authored under expert protocol. We evaluate frontier vision-language models (VLMs) with vendor-native and open-source harnesses. The best evaluated model achieves a score below 30\%, substantially lower than frontier performance on existing multimodal benchmarks. We further find that the choice of harness substantially affects model behavior, including but not limited to scores and failure modes. AgenticVBench provides a foundation for understanding and improving both models and harnesses for agentic video production.

Static bug analyzers play a crucial role in ensuring software quality. However, existing analyzers for bug detection in large-scale codebases often suffer from high false positive rates. This limitation largely stems from their inadequate capabilities in performing precise path feasibility validation for complex code contexts. While recent work has explored Large Language Models (LLMs) to eliminate false positives in static bug detection, the limited reasoning capabilities of LLMs over long and complex contexts hinder their effectiveness when applied to large-scale software projects. To address this challenge, we propose PathAgent, an agent-driven framework for fine-grained path feasibility analysis. PathAgent decomposes complex inter-procedural analysis into localized symbolic range reasoning sub-tasks, and employs an on-demand contextual exploration strategy to adaptively retrieve only the necessary code context. These designs enable precise path feasibility reasoning and effectively reduce false positives reported by static bug analyzers. Evaluation on large-scale real-world projects shows that PathAgent eliminates 79% of false positives, outperforming all baselines by 23% to 44%, while maintaining strong bug detection capability with a recall of 0.94.

Inverse Reinforcement Learning (IRL) is classically solved using Lagrangian optimization, by jointly optimizing a primal-dual problem with gradient descent to obtain the optimal reward function parameters and corresponding optimal policy. While this algorithmic view of IRL has been highly influential, gradient-based reward updates only paint a partial picture, largely ignoring the geometry of the problem in occupancy space. In this paper, we present an occupancy-space view of reward updates in IRL. We leverage the insight that both the optimal solution and its geometry in occupancy space are known, and show that the geometry along the e-geodesic (straight line in occupancy space) toward the optimal occupancy directly affects the gradient of the IRL dual problem in reward parameter space. This perspective naturally suggests preconditioning the dual gradient with the average Fisher information along the occupancy trajectory. We show that this preconditioned gradient is just a convex interpolation of the old reward and a new target reward, yielding efficient natural-gradient-style updates with little additional computational overhead. We then analyse these occupancy-space reward updates from a Mirror Descent (MD) perspective, and show that policy search over the resulting rewards induces approximate entropic MD on the policy. Finally, we conclude with convergence results of such geometry-aware IRL algorithms.

We study online learning in the adversarial injection model [GHMS23], where each instance is either drawn i.i.d. from an unknown distribution or is an arbitrary adversarial injection. The learner never knows which rounds are adversarial, but may abstain without penalty on adversarial injections. The learner's goal is to make few total misclassifications and few abstentions on i.i.d. instances. This model lies between classical statistical learning, governed by the VC dimension, and adversarial online learning, governed by the Littlestone dimension. Unlike prior work, which has largely focused on the realizable setting, we consider an agnostic formulation that allows arbitrary labels. We present an online algorithm that guarantees sublinear excess misclassification error and sublinear abstention error, against oblivious and adaptive adversaries, for function classes that admit bounded bracketing number witnessed by brackets of bounded VC dimension. Our algorithm is distribution-independent and does not require prior knowledge of the distribution. As corollaries, we obtain new learning guarantees for halfspaces (linear classifiers) and richer classes such as feed-forward neural networks with threshold activations.

We introduce HBAR (Hierarchical Brain Activity Representation), a hierarchical tokenization framework for learning compact, multi-scale representations of neuroimaging data.HBAR organizes voxel-level resting-state fMRI into a hierarchy spanning sub-parcel voxel groups, parcels, and large-scale functional networks, enabling approximately 415x compression while preserving fine-grained spatial information beyond conventional parcellation. The framework supports fixed atlas-based, fully learned, and softly prior-guided hierarchies, allowing us to systematically study the role of structured priors in brain representation learning. Evaluated on large-scale fMRI data, HBAR substantially improves reconstruction over parcellation-based and single-scale tokenization baselines, while supporting downstream modeling of brain dynamics and demographic prediction tasks such as sex classification. Across reconstruction and downstream evaluation, atlas-guided hierarchies provide the strongest overall performance, suggesting that established functional atlases encode organizational structure that remains difficult to recover from reconstruction alone. By combining voxel-level fidelity, compact discrete tokens, and multiscale brain structure, HBAR offers a practical framework for representation learning in high-dimensional neuroimaging data.

Learning neural solvers for spatiotemporal partial differential equations (PDEs) under real-world constraints remains a key challenge in scientific machine learning, especially for inverse tasks with sparse and noisy boundary observations. We present the Aletheia dataset, the first 3D benchmark for learning data-driven solvers in the context of nondestructive testing (NDT). The dataset simulates eddy-current-induced heating in conductive solids and models the resulting transient heat propagation governed by the heat equation. Aletheia contains over 4,700 high-resolution samples across 10 excitation frequencies (1-100\,kHz), each providing volumetric heat source and temperature fields over time. It supports both forward prediction of temperature evolution and inverse reconstruction of internal heat sources or defects from surface infrared measurements. Real infrared thermography data from cracked rail specimens are included for calibration and generalization studies. We define three canonical tasks on both regular and irregular grids and benchmark them using various neural operators. Aletheia establishes a unified platform for evaluating neural PDE solvers under realistic NDT conditions, enabling progress in reliable, data-driven inverse modeling.

Practical autonomous driving requires models that generalize by reasoning through spatial-temporal possibilities to exclude unsafe outcomes. While state-of-the-art (SOTA) methods use parallel planning architectures, they fail to explicitly couple speed decisions with agent behavior along the driving path, leading to suboptimal coordination. To address this, we propose a cascaded framework that transforms longitudinal planning from an independent prediction task into a path-conditioned reasoning process. On the model side, we introduce an anchor-based regression design that conditions longitudinal prediction on the lateral drive path, and reformulate longitudinal planning as 1D displacement prediction along the path. This reduces geometric uncertainty and sharpens the model's focus on interaction-driven dynamics. On the data side, we introduce a planning-oriented data augmentation strategy that simulates rare safety-critical events by programmatically inserting agents and relabeling longitudinal targets to enforce collision avoidance. Evaluated on the challenging Bench2Drive benchmark, our method achieves SOTA performance with a driving score of 89.07 and a success rate of 73.18\%, demonstrating significantly improved coordination and safety. Further evaluation on Fail2Drive confirms strong generalization to rare edge cases where parallel formulations typically fail. Visualization videos are provided in the supplementary material.

To advance the development of embodied intelligence, Open-Environment Active 3D Mapping has attracted increasing attention, aiming to perform a long-horizon and shortest trajectory exploration for reconstructing unseen scenarios. Since only limited information about unseen environments is available, methods built on the closed-set assumption, i.e., assuming that the test environments are similar to those seen during training, cannot generalize satisfactorily. In existing active mapping methods, long-horizon exploration is often guided by predicting a coarse long-range goal and then converting it into an executable path. However, this stage is usually formulated as single-point prediction. Under partial observability, the same local observation may correspond to multiple plausible exploration directions, making such deterministic prediction prone to brittle decisions and degraded performance in unseen scenarios. Our experiments further verify that this is a key factor underlying their weak generalization. To address this issue, we reformulate long-horizon target prediction as conditional multimodal anchor generation using Conditional Flow Matching. Instead of predicting a single goal, our method learns a conditional distribution over coarse exploration anchors from the current mapping state. These anchors are first converted into executable candidate paths through obstacle-aware planning. We then apply exploration-mode clustering to compress geometrically similar trajectories and reduce candidate redundancy. Finally, a hierarchical selection module selects the most promising mode and reranks paths within it to produce the final executable trajectory. Experiments show that our method improves generalization and reconstruction efficiency in open environments.

While the maximal coding rate reduction (MCR$^2$) objective has become a useful principle for learning compact and discriminative representations, a theoretical understanding of its optimization geometry remains limited, especially on why first-order methods often exhibit fast convergence near its local maximizers. In this work, we show that the MCR$^2$ objective satisfies an error-bound property in a neighborhood of its local maximizers. This property guarantees that the distance to the local maximizer set can be bounded above by the gradient norm at any point in the neighborhood, directly implying regularity conditions such as the Polyak-Lojasiewicz inequality and quadratic growth. Leveraging this property, we prove that first-order methods converge linearly to a local maximizer of the MCR$^2$ objective under mild conditions. Numerical experiments validate the predicted local convergence behavior and provide empirical evidence that optimizing the MCR$^2$ objective improves representation geometry while maintaining competitive downstream accuracy.

Extracting signals through alpha factor mining is a fundamental challenge in quantitative finance. Existing automated methods primarily follow two paradigms: Decoupled Factor Generation, which treats factor discovery as isolated events, and Iterative Factor Evolution, which focuses on local parent-child refinements. However, both paradigms lack a global structural view, often treating factor pools as unstructured collections or isolated single-lineage paths, which leads to redundant search and inefficient allocation of optimization effort. To address these limitations, we introduce AlphaPROBE, (Alpha Mining via Principled Retrieval and On-graph Biased Evolution), a framework that reframes alpha mining as the strategic navigation of a Directed Acyclic Graph (DAG). By modeling factors as nodes and evolutionary links as edges, AlphaPROBE treats the factor pool as a dynamic, interconnected ecosystem. The framework consists of two core components: a Bayesian Factor Retriever that identifies high-potential seeds by balancing exploitation and exploration through a posterior probability model, and a DAG-aware Factor Generator that leverages the full ancestral trace of factors to produce context-aware, non-redundant optimizations. Extensive experiments on major Chinese and US stock market datasets against 8 competitive baselines demonstrate that AlphaPROBE significantly gains enhanced performance in predictive accuracy, return stability and training efficiency. Our results confirm that leveraging global evolutionary topology is essential for efficient and robust automated alpha discovery. We have open-sourced our implementation at https://anonymous.4open.science/r/AlphaPROBE-A103.

Mixture-of-Experts (MoE) architectures scale model capacity through sparse expert activation, but their deployment remains memory-bound because all expert weights must reside in memory. Mixed-precision quantization can substantially reduce this footprint by assigning different bit-widths to different experts. Existing approaches, however, typically rely on calibration data to estimate expert importance and determine bit allocation. For frontier MoE LLMs, the original training data, and hence the true training distribution, is proprietary and inaccessible. As a result, calibration sets are inevitably imperfect surrogates, which can misestimate expert utilization and lead to suboptimal bit allocation. Motivated by the substantial cross-expert quality variability observed in modern MoE models, we propose AlphaQ, a calibration-free bit-allocation method for MoE quantization. AlphaQ draws on Heavy-Tailed Self-Regularization (HT-SR) theory and follows a simple principle: experts with more heavy-tailed weight spectra are typically better trained and should receive higher bit-widths, while experts with weaker heavy-tailed structure can be quantized more aggressively. AlphaQ operationalizes this principle by measuring expert-wise spectral heavy-tailedness and solving a budget-constrained optimization problem that minimizes total quantization error under a global bit-budget constraint. Across several MoE models, AlphaQ consistently outperforms calibration-based baselines under matched bit budgets. Notably, on Qwen1.5-MoE, AlphaQ achieves near full-precision accuracy with an average expert precision of only 3.5 bits, while delivering more than 4$\times$ memory compression.

While empirical scaling laws for LLM reasoning are well-documented, the theoretical mechanisms governing out-of-distribution (OOD) generalization remain elusive. We formalize reasoning via optimal transport, projecting discrete trajectories into a continuous metric space to quantify domain shifts using the Wasserstein-1 distance. Invoking Kantorovich duality, we bound OOD generalization via architectural Lipschitz continuity and functional approximation limits. This exposes two primary constraints. First, position-dependent attention (e.g., Absolute Positional Encoding) fails to preserve shift invariance, yielding an $\Omega(1)$ Lipschitz constant and expected risk, whereas shift-invariant mechanisms (e.g., Rotary Embeddings) preserve equivariance and bound the error. Second, by mapping sequential backtracking to a Dyck-$k$ language, we establish a strict circuit depth lower bound for $\text{TC}^0$ Transformers. Scaling physical layer depth is necessary to avert representation collapse---a constraint that scaling representation width cannot bypass due to irreducible approximation bounds in Barron spaces. Evaluations across 54 Transformer configurations on combinatorial search corroborate these bounds, demonstrating that generalization risk degrades monotonically with the Wasserstein domain shift.

Modern deep learning commonly relies on AdamW with prescribed learning rate schedules, but recent works challenge both components: Schedule-Free optimization removes explicit schedules via iterate averaging, and Muon improves the update geometry by orthogonalizing momentum for matrix parameters. Despite Muon's strong empirical performance, its underlying mechanism remains partially understood. We study Muon through the river-valley loss landscape, where useful training progress occurs along a flat, low-curvature bulk subspace (the river), while high-curvature dominant directions form steep valley walls that induce oscillations. We empirically show that while Muon's orthogonalization accelerates river progress by increasing the bulk component, it also amplifies dominant-direction noise, causing oscillatory trajectories. Building on this, we propose Anytime MUon with Stable gradient Evaluation (AMUSE), which integrates Muon's rapid bulk progress with the stabilizing effect of Schedule-Free averaging. AMUSE uses a time-varying interpolation coefficient that initially evaluates gradients near the fast Muon sequence for rapid adaptation, then gradually shifts toward the stable averaged sequence to suppress valley-wall oscillations. As a result, AMUSE requires no learning rate schedules and supports anytime training. Across vision tasks and large language model pretraining, AMUSE consistently improves the performance-iteration Pareto frontier over (Schedule-Free) AdamW and Muon.

Denoising generative models predict clean data ($x$), noise ($\epsilon$), or velocity ($v$), and are trained with a squared loss that can be computed in any of these three spaces. Prior work treats these choices as interchangeable up to loss reweighting, but practitioners observe sharp qualitative differences that remain unexplained. We systematically decouple the \emph{prediction space}, what the network outputs, from the \emph{loss space}, where supervision is applied, creating a $3{\times}3$ design matrix. We show that loss-space changes only reweight how much each noise level contributes to training, while prediction-space changes alter what the network is asked to learn: two fundamentally different mechanisms. This decoupling exposes two regimes. Without a representational bottleneck, the loss choice primarily controls denoising quality (MSE) while the prediction choice primarily controls sample quality (FID). The reason is that ODE sampling must convert the network's output back into a clean image, and this conversion amplifies errors differently across prediction types, turning sub-one-percent gaps in training error into 22-fold gaps in FID that are invisible to standard training diagnostics. Under a representational bottleneck, the prediction choice instead dominates both metrics: recovering noise or velocity requires reconstructing high-dimensional information that the bottleneck has destroyed, while recovering the clean image remains feasible because real images lie on a low-dimensional manifold. We validate these findings on a U-Net and the full 130M-parameter JiT architecture across CIFAR-10 and Imagenette, and give architecture-dependent recommendations for both regimes.

Central to human-aligned AI is understanding the benefits of human-elicited labels over synthetic alternatives. While human soft-labels improve calibration by capturing uncertainty, prior studies conflate these benefits with the implicit correction of mislabeled data (mode shifts), obscuring true effects of soft-labels. We present a controlled audit of soft-label learning across MNIST and a synthetic variant, re-annotating subsets to extract human uncertainty. By decoupling soft-label supervision from underlying label mode shifts, we show that while human soft-labels do provide accuracy gains, their larger value lies in acting as a regularizer that improves model calibration on difficult samples and promotes stable convergence across training runs. Dataset cartography reveals models trained on human soft-labels mirror human uncertainty, whereas those trained on synthetic labels fail to align with humans. Broadly, this work provides a diagnostic testbed for human-AI uncertainty alignment.

Unpaired medical image-to-image translation requires synthesizing target-modality images from source-modality images without aligned supervision. Existing methods learn a direct source-to-target mapping that entangles structure preservation and appearance generation within a single model, often distorting clinically meaningful anatomy in the translated output. We argue that these two objectives should be decoupled entirely: anatomical structure should be captured independently of generation, and the generator should operate only on a frozen structural representation. To this end, we propose a two-stage framework for unpaired medical image translation. In the first stage, we learn a shared structure space by training a structure encoder-decoder on source images, target images, and structural masks jointly, using mask reconstruction to anchor the latent space to modality-agnostic anatomy. For source images without mask annotations, a target-domain memory bank provides soft structural supervision via prototype retrieval. In the second stage, a conditional flow model learns to render target-domain appearance under fixed structural conditions. This ensures the generator inherits target appearance statistics and can focus entirely on structural conditioning rather than learning appearance from scratch. We validate our framework on OCT-to-OCTA synthesis and brain MR-to-CT translation. Our method outperforms all unpaired baselines on both tasks and, despite requiring no paired supervision, also surpasses paired baselines, improving PSNR by up to 11.2% and SSIM by up to 26.4% on OCT-to-OCTA, and achieving consistent gains across whole-image, soft-tissue, and bone regions on MR-to-CT.

Post-training methods for large language models (LLMs) implicitly aggregate diverse human feedback in ways that can be studied explicitly using social choice theory. In this paper, we adopt the perspective that for each post-training method, there is a corresponding reference-dependent probabilistic voting rule. Taking Direct Preference Optimization (DPO) and Nash Learning from Human Feedback (NLHF) as our case studies, we study axiomatic properties of the associated voting rules. First, we show that NLHF violates one of the central axioms of social choice, namely monotonicity, if and only if its KL regularization parameter is below a bound, whereas DPO satisfies monotonicity regardless of its regularization parameter. Second, we show that while DPO is often associated with Borda-style preference aggregation, when DPO is regarded as a probabilistic voting rule, it violates two variable population axioms that the standard probabilistic version of Borda satisfies. Third, we show that unlike DPO, NLHF violates a commutativity axiom from Bayesian epistemology, which has practical implications for staged post-training. Finally, we conduct an empirical study of the frequency and magnitude of axiom violations in the CoVal dataset using real base models.

We introduce the Thresholding Monte Carlo Tree Search problem, in which, given a tree $\mathcal{T}$ and a threshold $\theta$, a player must answer whether the root node value of $\mathcal{T}$ is at least $\theta$ or not. In the given tree, 'MAX' or 'MIN' is labeled on each internal node, and the value of a 'MAX'-labeled ('MIN'-labeled) internal node is the maximum (minimum) of its child values. The value of a leaf node is the mean reward of an unknown distribution, from which the player can sample rewards. For this problem, we develop a $\delta$-correct sequential sampling algorithm based on the Track-and-Stop strategy that has asymptotically optimal sample complexity. We show that a ratio-based modification of the D-Tracking arm-pulling strategy leads to a substantial improvement in empirical sample complexity, as well as reducing the per-round computational cost from linear to logarithmic in the number of arms.

Industrial anomaly detection is evolving from category-specific detectors toward unified models capable of handling multiple categories simultaneously. However, in multi-class setting, greater inter-category diversity not only compels enhanced reconstruction capacity to model normal patterns but also exacerbates the “identical shortcut” problem, wherein anomalies are likewise well reconstructed. Moreover, existing unified frameworks are typically confined to single-modal (RGB) inputs and lack multimodal capability. To address these, we propose an Efficient Cross-modal Feature Reconstruction (ECFR) model for unified multi-class anomaly detection, which harnesses the inherent difficulty of cross-modal reconstruction to alleviate the shortcut issue and amplify anomaly discrimination. The framework is built on two core modules: an Adaptive Feature Interaction and Recalibration (AFIR) module and a Hybrid Attention Convolution (HAC) module. Through feature interaction, fusion, and reconstruction, our model achieves two key outcomes: learning normal patterns of multi-class objects while forcing reconstruction failures for anomalous inputs. Extensive experiments on the MVTec 3D-AD and Eyecandies datasets validate our approach, which achieves state-of-the-art performance in multimodal multi-class anomaly detection while requiring significantly fewer model parameters (44.63 M) and lower computational complexity (15.77 GFLOPs) compared to previous multimodal anomaly detection methods.

Studies of human reasoning have shown that people are typically stronger at evaluating reasoning than producing it from scratch. In contrast, large reasoning models (LRMs) are trained to excel at producing long chains of reasoning to solve complex problems. How then do LRMs perform at evaluating reasons? We investigate this with the Valid-Answer-Invalid-Reasoning (VAIR) dataset: math problems and solutions with trivial reasoning flaws but valid answers, designed to isolate reasoning evaluation from the confound of reasoning production. Unlike humans, who we find are only 6\% worse at grading than solving such problems, we find a substantial production-evaluation gap in LRMs: frontier models score as low as 48\% when evaluating VAIR solutions, despite near-perfect solution production. Why this enigma? Through chain-of-thought (CoT) analysis, we find evidence of an answer confirmation bias: LRMs often produce then check for the correct answer instead of carefully verifying each step, fabricating rationalizations even when noticing anomalous reasoning. Linear probes corroborate this, showing that while LRM activations encode some representation of valid reasoning, they fail to robustly represent VAIR solutions as invalid. Causal patching of the final answer's representations causes LRM verdicts and activations to flip, demonstrating that answer validity is responsible for models' confirmation biases. These findings indicate an outstanding limitation in dominant approaches to reasoning training, which incentivize LRMs to produce and confirm reasoning towards correct answers, but not to robustly evaluate the underlying reasons.

Building black-box models for dynamical systems from data is a challenging problem in machine learning, especially when asymptotic stability guarantees are required. In this paper, we introduce a novel stability-ensuring and backpropagation-compatible projection scheme based on the Schur decomposition for the state matrix of linear discrete-time state-space layers, as well as an alternative pre-factorized formulation of the methodology. The proposed methods dynamically project the quasi-triangular factor of the state matrix's real Schur decomposition onto its nearest stable peer, ensuring stable dynamics with minimal overparameterization. Experiments on synthetic linear systems demonstrate that the method achieves accuracy and convergence rates comparable to those of state-of-the-art stable-system identification techniques, despite a marginal increase in computational complexity. Furthermore, the lower weight count facilitates convergence during training without sacrificing accuracy in stacked neural-network architectures with static nonlinearities targeting real-world datasets. These results suggest that the Schur-based projection provides a numerically robust framework for identifying complex dynamics on par with the State of the Art while satisfying strict asymptotic-stability requirements.

LLM-based systems are increasingly deployed in domains where incorrect outputs carry real costs, often necessitating expensive human review and verification. We argue that a central issue is not simply error, but consequence-insensitivity: models fail to adjust their behavior according to the cost of being wrong. We study this failure across both explicit utility framings as well as natural-language descriptions of stakes that mirror real-world deployment scenarios. Our evaluation spans agentic coding and mathematical reasoning, covering five frontier model families. Across settings, models systematically under-abstain: they continue to answer or submit patches even when incorrect answers carry significant consequences. The pathology is striking: models continue to submit answers in trivial settings where abstention is strictly dominant, and even when told an incorrect answer will cause nuclear extinction. Further, we find that this behavior is orthogonal to existing benchmarks; increasing model size or capability within a family does not improve sensitivity. Comparing base and instruction-tuned models localizes much of this anti-abstention bias to post-training: base models abstain far more often, though they are not themselves reliably consequence-aware. Finally, we evaluate prior post-hoc interventions alongside in-context learning and fine-tuning, finding that none robustly resolves the failure. Our results suggest that current post-training and evaluation pipelines optimize models to answer, not to act under stakes -- a significant bottleneck for trustworthy autonomy.

As wearable and mobile devices become increasingly embedded in daily life, they offer a practical way to continuously sense human motion in the wild. But inertial signals are highly dependent on the sensing setup, including body location, mounting position, sensor orientation, device hardware, and sampling protocol. This setup dependence makes it difficult to learn motion representations that transfer across devices and datasets, and limits the broader use of wearable IMUs beyond closed-set recognition. We introduce AnyMo, a geometry-aware framework for setup-agnostic human motion modeling. AnyMo uses physics-grounded IMU simulation over dense body-surface placements to generate diverse and plausible synthetic signals, pre-trains a graph encoder from paired synthetic placement views and masked partial observations, tokenizes multi-position IMU into full-body motion tokens, and aligns these tokens with an LLM for motion-language understanding. We evaluate AnyMo on three complementary tasks: zero-shot activity recognition across 14 unseen downstream datasets, cross-modal retrieval, and wearable IMU motion captioning, where it improves average Accuracy/F1/R@2 by 11.7\%/11.6\%/22.6\% on HAR, increases zero-shot IMU-to-text and text-to-IMU retrieval MRR by 15.9\% and 28.6\%, respectively, and improves zero-shot captioning BERT-F1 by 18.8\%. These results support AnyMo as a generalist model for wearable motion understanding in the wild. Code is at \url{https://anonymous.4open.science/r/anymo}.

Predicting high-fidelity flow fields around ship hulls is central to hydrodynamic performance analysis, yet traditional CFD solvers remain computationally prohibitive for rapid design exploration across multiple vessel types and operating conditions. Existing deep-learning surrogates often interpolate unstructured CFD meshes onto regular grids, which can blur near-field resolution, while direct point-level global attention on large unstructured meshes can be computationally expensive. In this paper, we introduce APPSolver, a neural surrogate that learns next-step point-wise flow-field prediction directly on the unstructured free-surface ship CFD mesh without re-gridding. APPSolver is built around Adaptive Patch Partitioning (APP), a quadtree-based strategy that groups unstructured points into local spatial patches according to the non-uniform density inherent in ship CFD meshes: fine near the hull and wake, coarse in the far field. The resulting patch tokens are processed by a Transformer backbone together with optional condition token that encode ship geometry and condition parameters via a frozen large language model, forming a unified token-space model for one-step temporal advancement of velocity and pressure fields. On the ShipBench dataset, APP-Transformer achieves the lowest MAE and MSE. Ablation studies quantify the compression-fidelity trade-off of APP and demonstrate that condition token provide preliminary gains in some leave-one-hull-out settings. The code will be released upon acceptance.

We propose a primal-dual policy optimization method for reinforcement learning in semi-infinitely constrained Markov decision processes (SICMDP), which extends standard constrained Markov decision processes by allowing a continuum of constraints. We call our proposed method \textbf{S}emi-\textbf{I}nfinitely \textbf{P}rimal-\textbf{D}ual \textbf{P}olicy \textbf{O}ptimization (SI-PDPO). By introducing an infinite-dimensional Lagrange multiplier, we derive the saddle-point formulation of the original SICMDP problem and prove strong duality under mild conditions. We then introduce a regularization term for the dual variable, and solve the resulting regularized saddle-point problem using first-order methods with a policy optimization subroutine (for example, NPG or PPO). Unlike existing primal-type policy optimization algorithms for semi-infinitely constrained reinforcement learning, our method does not require solving an inner-loop optimization problem, thereby mitigating computational intractability and reducing the over-conservative bias. We perform a series of numerical experiments, including a real-world power control task in wireless communications, to evaluate the performance of SI-PDPO. Empirical results show that SI-PDPO consistently outperforms existing primal-type algorithms.

Differential privacy (DP) has traditionally been used to provide theoretical upper bounds on an algorithm's stability to changing its training data. In modern machine learning applications, achieving strong tradeoffs between utility and theoretical privacy is challenging, and thus one may optimistically hope that existing theoretical analyses are loose. Recent work on \emph{privacy auditing} has adopted a dual viewpoint, instead lower bounding the true privacy of an algorithm by constructing empirical distinguishing events. The auditing literature has thus far yielded a pessimistic outlook on the looseness of theoretical privacy bounds for DP-SGD, the de facto private training method in modern ML, as nearly-matching empirical lower bounds have been achieved under various threat models \cite{NasrHSBTJCT23, AnnamalaiC24, CebereBP25}. In this work, we propose the empirical privacy lower bound of an algorithm as a concrete metric to optimize for, complementary to the theoretical upper bound. We give a lightweight defense framework that generically augments optimization methods in the ML pipeline to have significantly-improved empirical privacy on standard benchmarks. Moreover, we show that our framework comes at \emph{no theoretical privacy cost} when augmenting DP-SGD, unlike previous defenses against membership inference attacks. We evaluate our defense against a broad range of audit constructions, models, and datasets to demonstrate its flexibility. Our code implementation can be found at \href{https://github.com/neurips26privacyaudit/empirical-privacy-defense}{this anonymous repository}.

Large Language Models (LLMs) are increasingly used for automated feature engineering (FE) on tabular data, but prevailing benchmarks are ill-suited to evaluate this capability. They often assume i.i.d. samples, lack temporal dynamics and distribution shifts common in production, provide no known performance ceiling, and their artifacts (e.g., winning solutions) may enter LLM training corpora—blurring genuine discovery vs. memorization or spurious correlation. We introduce TabGen, a controllable synthetic generator that produces two synchronized views: an observable view for learners and an oracle view with minimal sufficient features, yielding a deterministic upper bound. TabGen composes non-i.i.d. dynamics to control inter-row interactions and distribution shift. Using TabGen, we evaluate several FE strategies, including Analyze&Act—a hypothesis-driven FE workflow proposed here. In no-drift settings, methods that instruct LLMs to iteratively analyze and refine data hypotheses consistently outperform alternatives, though LLM-driven FE shows substantial variance. Under drift, all methods degrade, and many fail to generalize—often underperforming the no-FE baseline—indicating FE-level overfitting to training-period correlations that do not transfer to future dynamics. Traditional i.i.d. benchmarks, by design, cannot reveal this failure mode.

Multimodal benchmarks are central to evaluating whether models can reason over information from multiple modalities. However, existing literature has revealed that some benchmarks may contain shortcut items that can be answered from a single modality or from answer-option regularities alone, thus failing to reflecting models' true cross-modal inference ability. Existing methods for assessing benchmarks typically rely either on intrinsic item structure or extrinsic model response behavior, which are not adequate for a comprehensive diagnosis. To fill in this gap, we propose SRCoD, a structure-response co-calibrated diagnosis framework for item-level diagnosis of modality dependence in multimodal benchmarks. SRCoD estimates sample-wise multimodal information structure from benchmark content and target answers, and uses this signal as a soft structural anchor for modality-conditioned item response modeling. The learned model produces a calibrated item-level diagnostic profile that reflects how strongly an item depends on different forms of modality evidence, enabling fine-grained benchmark analysis beyond aggregate accuracy. SRCoD also provides interpretable attribution of the diagnosis results. Extensive experiments on two complementary protocols, i.e., benchmark-refinement oriented validation and human-labeled item validation, demonstrate that SRCoD outperforms state-of-the-art benchmark diagnosis methods. Our code and data are available at https://anonymous.4open.science/r/SRCoD-61B0.

Current test-time scaling (TTS) techniques enhance large language model (LLM) performance by allocating additional computation at inference time, yet they remain insufficient for agentic settings, where actions directly interact with external environments and their effects can be irreversible and costly. We propose ARTIS, Agentic Risk-Aware Test-Time Scaling via Iterative Simulation, a framework that decouples exploration from commitment by enabling test-time exploration through simulated interactions prior to real-world execution. This design allows extending inference-time computation to improve action-level reliability and robustness without incurring environmental risk. We further show that naive LLM-based simulators struggle to capture rare but high-impact failure modes, substantially limiting their effectiveness for agentic decision making. To address this limitation, we introduce a risk-aware tool simulator that emphasizes fidelity on failure-inducing actions via targeted data generation and rebalanced training. Experiments on multi-turn and multi-step agentic benchmarks demonstrate that iterative simulation substantially improves agent reliability, and that risk-aware simulation is essential for consistently realizing these gains across models and tasks.

MRI morphometry pipelines are often evaluated by measurement agreement or downstream accuracy, yet their scientific value depends on the conclusions they support. This creates a gap: two pipelines may look similar at the feature level while supporting different conclusions about aging, cognitive decline, or neurodegeneration. To bridge this disparity, here we propose Scientific Claim Stability (SCS), a reusable claim-centric evaluation framework that converts region-of-interest (ROI) morphometry tables into interpretable claim cards and tests whether each claim remains supported under realistic perturbations. To demonstrate the efficacy of SCS, we evaluate it on 10,370 MRI scans from three datasets and five cohorts, generating 2,783 claim cards and evaluating 89 claim-family pairs across dataset, pipeline, atlas, feature-family, and target shifts. SCS identifies claims that are stable, transferable, or fragile, and reveals high-concordance fragile-support cases where evidence vectors remain similar while supporting regions change. To assess claim plausibility, clarity, faithfulness, usefulness, and caution level, we further implement and compare an expert (including geriatrics, neuroscience, and radiology) review workflow and an LLM (including ChatGPT and Gemini) review workflow. Our results suggest claim-level evaluation as a practical standard for auditing MRI morphometry pipelines without requiring a universal gold standard.

Preference-conditioned multi-objective reinforcement learning aims to learn a single policy that captures trade-offs across preferences, but under nonlinear scalarization the uniqueness and continuity of the preference-to-solution correspondence remain unclear. We study this problem in tabular multi-objective Markov decision processes (MDPs) using smooth Tchebycheff scalarization as a monotone utility. Under mild interior conditions on the preference set, we prove that each preference induces a unique Pareto-optimal return vector and that this vector depends Lipschitz-continuously on the preference, providing a principled foundation for preference sweeping toward dense Pareto-front coverage. To compute these targets, we formulate the problem over occupancy measures and derive Concave Mirror Descent Policy Iteration (CMDPI), which achieves an $O(1/k)$ objective-suboptimality rate. We further show that each update is equivalent to solving a Kullback-Leibler-regularized MDP with the previous policy as reference, yielding a policy-iteration interpretation and finite-iterate policy continuity across preferences. We instantiate the update as a deep actor-critic algorithm preserving previous-policy regularization. On eight MO-Gymnasium tasks, it achieves the best average hypervolume rank among recent baselines and strong expected-utility performance. Continuous-control experiments indicate gains beyond the discrete-action setting.

The Objective Structured Clinical Examination (OSCE) is widely used to assess clinical communication skills in medical education, requiring trainees to take histories and reason clinically with standardized patients (SPs). Large language model (LLM)-based virtual standardized patients (VSPs) offer a practical alternative to human SPs. These are typically built using persona-injected prompting, which casts a model into a clinical role via a single system prompt. However, existing evaluations of such VSPs often rely on a single backbone LLM, small human rater panels, or LLM-as-a-judge scoring without an utterance-aligned human-expert reference. Consequently, they are limited in their ability to identify the specific shortcomings of individual models. In this study, we evaluated four LLMs (GPT-4o, Claude 3.5 Sonnet, Llama-3.1 70B, and HyperCLOVA X)—spanning commercial, open-source, and Korean-developed models—against a human-expert reference across three Korean OSCE cases (hematuria, fever in pregnancy, and chronic cough). In a double-blind, fully crossed design, 17 senior medical-student raters scored every model utterance on three sentence-level and six encounter-level Likert metrics, yielding 7,480 data points. We analyzed the scores using a stratified linear mixed-effects model based on an 11-category history-taking and 3-level linguistic-form taxonomy. At the encounter level, three of the four LLMs matched the human-expert reference on Consistency and Comprehensibility. HyperCLOVA X additionally matched the reference on Engagement, whereas Llama-3.1 70B underperformed the human reference across all six metrics. Furthermore, we introduce a web-based OSCE platform that supports virtual practice across diverse clinical cases. This platform continuously collects student–VSP dialogues under research and privacy consent; the resulting dataset is available at https://huggingface.co/datasets/neurips-2026-cpx/neurips-2026-cpx.

Standard Message Passing Neural Networks (MPNNs) are at most 1-WL expressive, which has spurred the development of higher-order models. In particular, many higher-order designs rely on extra global structural features (e.g., positional encodings or pairwise distances), whereas the $k$-WL hierarchy offers a principled and quantifiable path to increased expressiveness. However, directly simulating $k$-WL is computationally prohibitive, and existing $k$-WL variants often sacrifice expressiveness to gain efficiency, failing to capture critical structural distinctions. To overcome these shortcomings, we propose a higher-order graph learning framework based on walk-induced lifted states with structure-aware state encoding. We further design a controllably sparsified and localized $k$-FWL-style aggregation scheme, enabling graph-level higher-order representations via simple message passing. On the recent and challenging BREC expressiveness benchmark, our model achieves state-of-the-art total distinguishing accuracy among compared higher-order GNN baselines, outperforming strong 3-WL baselines, and remains competitive on real-world graph classification tasks.

Material synthesis planning (MSP) remains a fundamental and underexplored bottleneck in AI-driven materials discovery, as it requires not only identifying suitable precursor materials but also designing coherent sequences of synthesis operations to realize a target material. Although several AI-based approaches have been proposed to address isolated subtasks of MSP, a methodology for solving the entire MSP task has yet to be established. We propose MSP-LLM, a structured LLM-based framework that formulates MSP as a two-stage process composed of two constituent subproblems: precursor prediction (PP) and synthesis operation prediction (SOP). Our approach introduces a discrete material class as an intermediate decision variable that organizes both tasks into a chemically consistent decision chain. For SOP, we further incorporate hierarchical precursor types as synthesis-relevant inductive biases and employ precursor constraint factorization that preserves precursor-related information in the autoregressive decoding state. Extensive experiments show that MSP-LLM consistently outperforms existing methods on both PP and SOP, as well as on the MSP task, demonstrating an effective and scalable framework with practical potential for autonomous MSP.

Vision-language models (VLMs) are increasingly proposed as general-purpose tools for scientific data interpretation, yet their reliability on real astronomical observations across diverse modalities remains untested. We present AstroVLBench, a benchmark of over 4,100 expert-verified instances across five tasks spanning optical imaging, radio interferometry, multi-wavelength photometry, time-domain light curves, and optical spectroscopy. We systematically evaluate six state-of-the-art frontier models and find that performance is strongly modality-dependent and uncorrelated with general benchmark rankings. Mechanistic ablations reveal three concrete bottlenecks. First, prompts that ground attention in physical mechanisms (why a feature matters) yield more balanced classifications than phenomenological prompts (what to look for). Second, presenting one-dimensional measurements as numerical tables rather than rendered plots improves accuracy by up to 13 percentage points, indicating that plot rendering can obscure physically relevant signal. Third, a qualitative reasoning analysis reveals a pervasive “right-answer-wrong-reason” phenomenon: models routinely reach correct predictions through physically invalid logic, showing that accuracy alone is insufficient for trustworthy scientific deployment. AstroVLBench provides the first systematic, multi-modal baselines for VLMs in observational astronomy and identifies the representation, grounding, and reasoning bottlenecks that must be addressed before VLMs can be trusted as autonomous scientific interpreters.

Spatial cognition requires stable neural manifolds of orientation and location that track self-motion and sensory cues. Continuous attractors provide a natural dynamical principle, but learning such mechanisms that form and control manifolds with theoretical guarantees remains difficult. We propose Spatial Energy Attractor Learning (SEAL), a theory of neurodynamics learning for spatial cognition, where continuous spatial representations arise as learned low-energy manifolds in Hopfield energy landscapes through population-level Fourier energy learning. For head-direction and grid-cell systems, Fourier energy landscapes yield ring and torus attractors with learned normal attraction and input-driven tangent transport. We prove that Fourier construction is a special case of a general Hopfield-compatible energy extension whose autonomous dynamics preserve energy descent and whose expressive energy families approximate target attractor energies and restoring fields. Experiments on HD and grid-cell populations show that the learned energies restore perturbed states to ring and torus manifolds, stabilize velocity-driven moving bumps across multiple input regimes, and induce continuous-attractor interactive structures with local excitation and surround inhibition. Together, these results provide a stable, learnable, and biologically interpretable energy-based account of continuous spatial representation.

Predictive Vision–Language–Action (VLA) models aim to improve robotic manipulation via future observation or world dynamics forecasting. However, existing approaches often fail to realize this potential and underperform direct action prediction models. We argue that these limitations stem from modality misalignment between observations and actions, together with joint optimization conflicts that drive learning away from an action-centric objective. To this end, we introduce ATI-VLA, an Action-Centric Predictive Vision–Language–Action framework via Actionable Alignment Then Adaptive Injection. Specifically, it follows a two-step design: 1) Actionable Representation Alignment via a Shared Codebook. It aligns predictive observation and action representations by mapping both modalities into a shared discrete latent space via a unified codebook, making predictive observation latents readily usable for action generation and mitigating modality misalignment. 2) Action-Centric Adaptive Injection of Predictive Latents. Building upon this, it then injects predictive observation latents into action decoding as explicit predictive priors via a lightweight adaptive side-path, enabling adaptive predictive guidance under a single action-centric objective. Extensive experiments on both simulation and real-world robotic tasks demonstrate that ATI-VLA achieves state-of-the-art performance with faster convergence.

High-fidelity atomistic evolution over long timescales requires more than observing the current crystal configuration. Instantaneous atomistic snapshots are often incomplete: locally similar configurations can correspond to different hidden dynamical contexts, future event preferences, and waiting-time scales. We argue that this snapshot ambiguity makes long-horizon atomistic evolution fundamentally a memory-based world-state restoration problem. To address this, we introduce AtomWorld-Mem, a memory-restored atomistic world model that recovers the latent world state missing from instantaneous crystal snapshots. AtomWorld-Mem treats the evolving alloy as an AtomWorld: spatial encoders write multi-scale atomistic keyframes from dense local topology and sparse long-range defect context, while short-term event memory and long-term structural memory integrate these keyframes across time to restore a future-predictive evolutionary state. The restored state is used to prioritize legal vacancy-mediated events under single-event Kinetic Monte Carlo (KMC) constraints, while event legality, physical execution, and residence-time updates remain governed by the underlying simulator. Empirically, AtomWorld-Mem improves long-horizon atomistic progress under fixed microscopic event budgets while maintaining high-fidelity evolution across energetic, structural, and vacancy-transport observables. It further transfers zero-shot across diverse unseen alloy-temperature AtomWorlds, suggesting that the learned memory-restoration mechanism captures reusable principles of hidden-state inference rather than a system-specific local energy heuristic. These results position memory-restored world-state modeling as a promising route toward efficient, physically grounded, and transferable atomistic evolution.

Robust visual classification often depends on localizing the main foreground objects in an image while ignoring spatially-separable distractors. Surprisingly, we find that the attention maps of smaller self-supervised ViTs localize foreground objects better than those of larger ones. However, we still need large ViTs, because they extract richer representations from each patch. To get the best of both worlds, good localization _and_ rich representations, we propose $A^2$, a simple method that leverages this inverse scaling finding by decoupling _where to look_ (a small attention model) from _what to extract_ (a large embedding model): we crop around the attention peaks of a small model and embed the crops with a larger model. $A^2$ uses entirely pretrained features and does not require per-dataset attention or backbone training. Across $5$ benchmarks, $A^2$ is competitive with backbone-matched loss-level methods like DFR, and outperforms end-to-end attention training under stronger distribution shifts.

Understanding how the brain integrates information across sensory modalities under naturalistic settings remains a central challenge in cognitive computational neuroscience. Despite recent progress in multimodal brain encoding driven by the advancement of deep learning, many existing approaches combine representations from pretrained modality-specific models using static aggregation or linear readouts, and therefore do not explicitly capture how information is selectively integrated across cortical regions. Here, we introduce Multimodal Transformer Brain Encoder (M-TBEn), a neural encoding framework that models multimodal integration as a routing problem. M-TBEn extends prior vision-based brain encoding approaches by introducing a cross-attention mechanism for multimodality, in which learnable parcel-level queries dynamically select and aggregate modality-specific representations from visual, auditory, and linguistic inputs. This design is not only parameter-efficient but also yields parcel-specific multimodal representations that support a flexible linear readout for predicting brain responses at various spatial resolutions. We evaluate M-TBEn on naturalistic video-viewing data, including both in-distribution and out-of-distribution stimulus conditions. The model is able to accurately predict parcel-level and voxel-level fMRI responses and generalizes across stimulus distributions. Furthermore, the learned attention patterns provide a structured descriptive basis for analyzing modality integration across cortical parcels, enabling a connection between parcel-specific routing profiles and established functional brain networks. Together, these results suggest that attention–based routing offers a principled computational framework for modeling multimodal integration in the human brain.

Visual Geometry Grounded Transformer (VGGT) advances 3D reconstruction via scalable Transformer architecture, but the quadratic complexity of global attention prevents long context application. StreamVGGT enables streaming with causal attention, yet its KV cache grows linearly with frames, causing memory overflow and quality degradation. We present RetrieveVGGT, a training-free framework, which formulates context construction for VGGT as a retrieval problem. By retrieving a fixed number of relevant frames at each step, VGGT maintains a controllable memory budget, which is close to its training context length. Interestingly, we find that the similarity between current frame queries and cached history frame keys at the first global attention layer of VGGT is already a strong indicator of relevance, eliminating the need for additional learned scoring. To enhance information diversity similar to a recommender system, we propose Segment Sampling so that the retrieval spans distinct relevant segments rather than a single high-similarity region. We design a pose-aware spatial memory mechanism that organizes history frames according to their already estimated camera poses, enabling location-aware retrieval. Extensive experiments demonstrate that RetrieveVGGT achieves state-of-the-art performance, outperforming StreamVGGT, TTT3R, and InfiniteVGGT while maintaining constant memory usage regardless of sequence length.

Transformer attention exhibits attention sinks, where a small subset of source tokens absorbs disproportionate attention mass across layers. Recent work shows that post-SDPA output gating strongly reduces attention-sink behavior, while separate studies link attention sinks to gradient sinks and massive activations through backward training dynamics. Yet how the same gate affects forward sink structure, backward gradient concentration, and training-time representation signatures remains unclear. We address this gap by formulating attention sinks as sink-induced low-rank modes in token-space covariance, with directions aligned to high-mass attention columns. Under this view, post-SDPA gating acts as a mode-dependent contraction: it attenuates sink-induced spectral energy in the forward pass and locally reduces value-path gradients routed through sink columns in the backward pass. Empirically, random-matrix diagnostics show that gating reduces outlier mass, weakens sink-subspace alignment, and increases effective rank, while backward analyses show active-gate attenuation of first-token value gradients. Training-time diagnostics further show that gating suppresses the co-emergence of attention concentration, value-gradient amplification, and representation dominance.

A recent work shows that Attention Transfer, which transfers only the attention patterns from a pre-trained teacher Vision Transformer (ViT) to a randomly initialized standard student ViT, is sufficient to recover the full benefit of the teacher's pre-trained weights. We revisit this finding on a comprehensive benchmark of 20 teachers from 11 well-known ViT families and reveal that Attention Transfer is not universally effective. While 7 families transfer successfully, 4 consistently fail, falling up to 5.1\% below the from-scratch no-transfer baseline. Further results demonstrate that this failure is family-consistent across model sizes, and persists under extended training durations, different transfer datasets, and out-of-distribution evaluations. Controlled analyses then consistently localize the problem to the attention-routing channel, indicating that the key issue is not whether the student can match the teacher's attention patterns, but whether the matched patterns remain functional for the student. Crucially, we identify architectural mismatch between the pre-trained teacher and the standard student as the primary mechanism. By adding only the teacher's native architectural components to the student in a randomly initialized state, we completely reverse the failure for all 4 families. Notably, these components alone do not improve from-scratch training, confirming that they specifically unlock the usability of the teacher's attention. We further systematically show that this failure is not explained by the inadequate choice of transfer loss or by differences in pre-training recipes. Our findings refine the prevailing understanding of attention in ViT representations: attention is sufficient only when the student architecture matches the teacher.

Vision-Language-Action (VLA) models have achieved remarkable success in robotic manipulation. However, their robustness to instruction variations remains a critical, under-explored safety concern, posing a significant safety risk to real-world deployment. Red teaming, or identifying environmental scenarios that elicit catastrophic behaviors, is an important step in ensuring the safe deployment of embodied AI agents. Reinforcement learning (RL) has emerged as a promising approach in automated red teaming that aims to uncover these vulnerabilities. However, standard RL-based adversaries often suffer from severe mode collapse due to their reward-maximizing nature, which tends to converge to a narrow set of trivial or repetitive failure patterns, failing to reveal the comprehensive landscape of meaningful risks. To bridge this gap, we propose a novel \textbf{D}iversity-\textbf{A}ware \textbf{E}mbodied \textbf{R}ed \textbf{T}eaming (\textbf{DAERT}) framework, to audit VLA robustness under semantically aligned instruction. Our design uses a breadth-seeking value estimator that prevents the attacker from collapsing onto a single high-reward phrasing, generating a diverse set of challenging instructions while preserving attack effectiveness, measured by execution failures in a physical simulator. We conduct extensive experiments across different robotic benchmarks against two state-of-the-art VLAs, including $\pi_0$ and OpenVLA. Our method consistently discovers a wider range of more effective adversarial instructions that reduce the average task success rate from 93.33\% to 5.85\%, demonstrating a scalable approach to stress-testing VLA agents and exposing critical safety blind spots before real-world deployment.

Anatomical mesh segmentation requires models that operate directly on irregular surface geometry while remaining robust to arbitrary patient pose and mesh resolution variation. Existing task-specific mesh and point-cloud methods are not equivariant, and can degrade sharply under test-time perturbation, for example dropping by 25-26 IoU points on intraoral scan segmentation at 40$\textdegree{}$ tilt. We present EAMS, an Equivariant Anatomical Mesh Segmentor built on Equivariant Mesh Neural Networks (EMNN), and evaluate it across four clinically distinct tasks spanning edge-, vertex-, and face-level supervision. We combine intrinsic mesh descriptors with anatomy-aware priors, including PCA-derived frames for dental arches and liver surfaces, and augment message passing to provide lightweight global context. Across intracranial aneurysm and intraoral segmentation, EAMS variants are competitive with specialized baselines on unperturbed inputs while remaining stable under geometric perturbations, and on liver surfaces they expose a favorable trade-off between canonical-pose accuracy and rotation robustness. These results show that a lightweight ($<2$M parameters) equivariant framework can deliver robust anatomical mesh segmentation across diverse supervision types without task-specific architectures.

Anchored fixed point and monotone equation methods, including Halpern iteration, extra anchored gradient, and their relatives, add a vanishing pull toward a reference point to obtain last-iterate guarantees. Existing anchored variants often achieve sharp last-iterate guarantees, but from the update-level perspective the placement of the anchor can be algorithm-specific and conceptually opaque. We show that anchoring admits a single operator-side construction: regularize the operator queried by the base method with a vanishing Tikhonov term, then run the unmodified base method. Applied to the Picard iteration, this recipe reproduces the Halpern iteration; applied to the forward step, extragradient, and Popov methods, it yields three variants whose anchor placements inherit the base method's query pattern. The four analyses share a residual recurrence, recovering the (O(1/k)) Halpern residual-norm convergence rate, giving (O(1/\sqrt{k})) for the regularized forward step, and giving (O(1/k)) for the regularized extragradient and Popov variants in the unconstrained monotone Lipschitz setting.

Post-training data pipelines are traditionally hand-designed by researchers who orchestrate generation, verification, and diversity balancing. While Large Language Models (LLMs) can already execute individual subtasks, the architectural composition of these pipelines remains a manual bottleneck, excluding the most substantial engineering step in post-training from automation. We introduce AutoDataBench, a benchmark evaluating whether LLM agents can autonomously design post-training data-synthesis pipelines. Each task gives the agent only a description of a target capability and asks it to produce a complete, executable pipeline whose generated corpus is then used to fine-tune a fixed student model and scored on the evaluation. Each task is delivered as a natural-language instruction and a sandboxed workspace given to the agent, paired with an evaluation protocol. We instantiate AutoDataBench on five capabilities: instruction following, competition math, deep search, repo-level software engineering, and terminal automation. For each task we pair the evaluation with a strong human-designed pipeline whose downstream score serves as a reference point for the agent. Experimental results reveal that frontier agents close most of the gap to the human reference on text-centric tasks (e.g., Opus-4.7 nearly matches humans in instruction following), yet a noticeable performance gap persists in tasks requiring executable-environment construction (SWE and Terminal). Furthermore, we observe a decoupling between downstream task-solving strength and pipeline-construction quality, where strong reasoning models like GPT-5.5 may still produce suboptimal pipelines. These findings suggest that autonomous environment synthesis remains the primary bottleneck for self-improving data loops, distinct from mere task-solving competence.

Manifold embedding for data visualisation is dominated by a small set of hand-designed state-of-the-art (SOTA) algorithms. Their design targets are fixed at development time, resulting in inflexible algorithmic behaviour that cannot be readily steered toward user-specified structural priorities. Hand-designing new embedding algorithms to address user priority requires highly specialised expertise and is time consuming. To address autonomous algorithm design tailored to user preference, we introduce an agentic algorithm generation pipeline AutoManifold. It composes new manifold embedding algorithms from a constrained vocabulary of affinity, cost, and optimisation primitives, supported by multi-agent large-language-model (LLM) orchestration. AutoManifold conditions every stage of its design on user-specified structural preservation preferences, and iteratively refines the algorithm configuration through an LLM-guided, metric-grounded iterative loop. We compare the generated algorithms against three strongest and most frequently used SOTA (t-SNE, UMAP, and PaCMAP) on real-world datasets spanning different difficulty regimes under identical evaluation infrastructure. The results demonstrate that LLM-driven, priority-conditioned algorithm synthesis can move beyond hyperparameter tuning, producing genuinely new algorithms that outperform traditional hand-designed methods, in their ability to steer towards user-specified properties without compromising much inherent structure in the original data.

We study the fundamental and timely problem of learning long sequences in autoregressive modeling and next-token prediction under model misspecification, measured by the joint Kullback--Leibler (KL) divergence. Our goal is to characterize how the sequence horizon (H) affects both approximation and estimation errors in this joint-distribution, sequence-level regime. By establishing matching upper and lower bounds, we provide, to our knowledge, the first complete characterization of long-horizon error behavior under the natural joint KL objective, with improved rates and optimality justification relative to existing work. On the approximation side, we show that joint KL admits a horizon-free approximation factor, in sharp contrast to Hellinger-based analyses that exhibit an (\Omega(H)) dependence for computationally efficient methods; this isolates the choice of divergence as the source of approximation amplification. On the estimation side, we prove a fundamental information-theoretic lower bound of order (\Omega(H)) that holds for both decomposable policy classes and fully shared policies, matching the (\widetilde O(H)) upper bounds achieved by computationally efficient algorithms. Our analysis clarifies the landscape of recent autoregressive learning results by aligning the log-loss training objective, the sequence-level evaluation metric, and the approximation metric through a sharp joint-KL oracle theory. We further show that these joint-KL guarantees imply policy learning regret bounds at rates matching prior imitation learning literature.

Hallucinations remain a major obstacle to deploying large language models (LLMs) in knowledge-intensive settings, where generated responses must be faithfully grounded in provided evidence. Reinforcement learning (RL) is a promising direction for hallucination mitigation, but response-level faithfulness rewards suffer from a granularity mismatch: localized hallucinations can cause supported content to receive spurious penalties. Although recent work introduces fine-grained feedback such as claim-level verification and token-level rewards, unbalanced credit assignment can still induce length, verbosity, or optimization-noise biases. We propose $\textbf{BALTO}$, a $\textbf{Bal}$anced $\textbf{T}$oken-level Policy $\textbf{O}$ptimization framework for hallucination mitigation. BALTO extracts checkable factual claims, verifies them against the reference context, and projects claim-level judgments to token-level labels. A balanced token-level credit assignment mechanism is introduced into the framework. This design redistributes probability mass from unsupported content toward faithful content, rather than suppressing the entire response. We systematically analyze the limitations of response-level rewards from a theoretical standpoint, and prove BALTO’s advantages in training stability and optimization efficiency for hallucination mitigation. Experiments on ConFiQA, RAGTruth, and FinLLM-Eval show that BALTO achieves the highest faithfulness across all six model--benchmark settings and consistently outperforms existing post-training baselines in Q-Score, demonstrating a stronger faithfulness--informativeness trade-off.

Current open-source diffusion models struggle to generate stable and synchronized audio-visual content, particularly in scenarios demanding complex semantic reasoning. The root cause is that existing methods rely on coarse text embeddings from off-the-shelf encoders to guide audio-video denoising, which discards fine-grained semantics and, critically, lacks a shared long-horizon plan, leading to uncoordinated denoising trajectories and fragile cross-modal alignment. We propose Baton, the first framework that introduces explicit semantic planning into joint video-audio generation. Our key insight is that complementing coarse text guidance with semantically rich, modality-aware planned tokens, jointly reasoned and mutually aligned before denoising, can simultaneously restore fine-grained semantic detail and establish a shared blueprint that coordinates both audio and video denoising trajectories. Concretely, Baton first introduces the VA-Planner, a multimodal language model equipped with dual semantic alignment towers, where learnable queries cross-attend to both video and audio features to produce a pair of semantically aligned video and audio planning tokens as keyframe-level blueprints. These planning tokens are injected into the diffusion backbone via cross-attention layers, providing temporally grounded guidance complementary to coarse text embeddings. Since planning tokens do not share one-to-one spatial-temporal correspondence with diffusion latents, we further propose Relative Semantic RoPE, a relative positional encoding that maps planning tokens and latents into a shared spatial-temporal coordinate frame, enabling each latent to accurately attend to its positionally corresponding semantic cues. Experiments on benchmarks show the effectiveness of Baton both qualitatively and quantitatively.

Modern ML systems increasingly use uncertainty to decide when to abstain, route to a fallback, retrieve more context, or spend additional computation. Yet many strong Bayesian deep-learning recipes make uncertainty expensive on the serving path: ensembles, posterior samples, and MC dropout require repeated forward passes, while fast deterministic confidence heads are not generally posterior-predictive computations over parameter beliefs. We introduce \emph{Bayesify}, a message-passing view that turns a deterministic neural computation graph into a single-pass posterior-predictive moment graph. Trainable tensors become Gaussian beliefs, activations carry selected moments, and prediction is one deterministic forward propagation of those moments. Theoretically, we show that exact belief propagation is natural-gradient message passing in exponential-family mean space, and that projected EP is implemented by reverse-mode adjoints followed by local Fisher moment updates. This yields a practical wrapper for dense, convolutional, residual, and adapter modules. Experiments evaluate calibrated uncertainty, selective prediction, and OOD behavior at one-pass inference cost, comparing against deterministic calibration, ensembles, dropout, Laplace, SWAG, and variational baselines.

Self-Driving Chemical Laboratories (SDCLs) are moving chemical experimentation toward embodied automation, where machines must perceive laboratory scenes, reason about experimental states, and act under physical, procedural, and safety constraints. Although Multimodal Large Language Models (MLLMs) are increasingly considered as embodied brains for such systems, existing laboratory benchmarks mainly focus on safety, anomaly detection, or general scientific understanding, leaving their execution-oriented embodied capabilities underexplored. To fill this gap, we introduce BEAKER, a Benchmark for evaluating Embodied Actionability, Knowledge, Experimentation, and Reasoning in SDCLs. BEAKER contains 1,000 expert-curated QA samples across 11 image and video task types, covering key judgments required for chemical laboratory execution, including affordance grounding, trajectory planning, equipment and state understanding, apparatus connection, operation process reasoning, and result understanding. All questions and annotations are manually created and reviewed by chemistry and embodied AI experts, without AI-assisted generation or annotation. We evaluate 28 general-purpose, scientific, chemical, and embodied MLLMs on BEAKER. Results show that current models can recognize laboratory objects and states, but still struggle to ground experimental knowledge into operable regions, feasible trajectories, and long-horizon procedural reasoning. BEAKER thus reveals a clear gap between multimodal laboratory perception and action-grounded experimental understanding, providing a diagnostic benchmark for future MLLMs in SDCLs.

Behavioral Foundation Models (BFMs) are an emerging paradigm in reinforcement learning, playing a role analogous to large language models in natural language processing: they have shown remarkable versatility, enabling zero-shot performance, fast imitation, and online adaptation, all by exploiting the structure of a latent space. In this work, we investigate whether the latent behavioral space induced by BFMs can serve as an effective search space to discover large repertoires of behaviorally diverse and high-performing policies through Quality-Diversity (QD) methods. While QD methods generally search directly in high-dimensional policy parameter space, in this paper, we present BFM-QD, a framework that performs QD search in the compact latent space of a BFM. We further show that the BFM-QD framework provides a closed-form, gradient-free policy improvement operator that approximates a policy gradient update, but requires no critic training and no backpropagation. Across continuous-control benchmarks spanning dense locomotion, sparse navigation, and contact-rich manipulation, BFM-QD consistently outperforms parameter-space baselines, with particularly stark gains in sparse and deceptive settings, where all existing QD methods collapse to near-zero performance. These results show the effectiveness of the BFM-QD framework, benefiting from the synergy between dimensionality reduction of the search space and offline pretraining from diverse behavioral data. This positions BFMs as a general-purpose backbone for QD optimization, extending their utility beyond zero-shot task solving to the discovery of diverse behavioral repertoires.

Imitation Learning (IL) is a natural framework for learning in sequential decision-making systems and has emerged as the dominant paradigm through which we understand language model training. A central puzzle is that, while in theory offline IL can be horizon-free and optimal, in practice online methods such as on-policy distillation (OPD) often outperform offline methods such as supervised fine-tuning (SFT). We propose a noisy expert model to explain this gap, in which the learner only has access to a noisy version of the expert's policy, but wishes to compete against the reward achieved by a clean expert, motivated by the fact that in many applications, e.g. training language models to perform long chains of thought, the expert is often imperfect. In this setting, we show a sharp separation between offline and online IL. Offline learning from noisy trajectories is fundamentally hard: to compete with the clean expert, the sample complexity must grow exponentially, in contradistinction to the clean expert setting where no explicit horizon dependence exists. In contrast, we prove that online interaction with the noisy expert via a novel variant of OPD enables horizon-free guarantees in some settings and polynomial dependence on horizon in general. Our analysis leads to an alternative loss function form that is commonly considered empirically for LM training. We further provide algorithms and lower bounds, and extend our results to the more realistic setting of unknown corruption when the clean expert is deterministic, thereby providing a theoretical foundation for why on-policy distillation can outperform standard supervised fine-tuning when training language models from imperfect teachers. We complement our theoretical results with experiments on synthetic and natural-language tasks, showing that the OPD variant suggested by our theory outperforms both offline BC and existing OPD objectives under noisy expert feedback.

Video multimodal large language models (MLLMs) keep climbing video question answering benchmarks, yet shuffling the frames, masking the segment that supports the answer, or occluding the target object barely changes their predictions. The accuracy rests on appearance and language priors, not on the temporal evidence the question asks for. We trace this to the unit of post-training: rewards are computed on a single response to the original clip, so the model is never asked to behave consistently across views. We propose Behavior Pack Optimization (BPO), which replaces the single response with a behavior pack of outputs across counterfactual views chosen by question type, scored jointly. The pack reward asks for stability when the intervention is irrelevant, sensitivity when key evidence is removed, and abstention when no evidence remains. To keep this objective stable at small pack sizes, BPO uses an anchor-relative advantage: the response on the original view serves as a per-prompt reference instead of a group mean over mixed views. On TempCompass, MVBench, and NExT-QA, BPO improves the macro accuracy of Qwen2.5-VL-7B-Instruct by 4.7 pp, the temporal-hard subset by 7.8 pp, and abstention F1 by 20.0 pp over a budget-matched vanilla GRPO baseline from the same SFT checkpoint. The gains transfer to Video-MME, LongVideoBench, and to LLaVA-Video-7B; ablations confirm they follow the view sets, not the rollout count. We hope this pack-level perspective offers a useful starting point for the video MLLM and multimodal post-training community as the field moves toward evidence-grounded video reasoning.

The rapid advancement of large language models (LLMs) has expanded their potential in the legal domain. However, existing legal benchmarks remain largely English-centric and oriented toward common law, leaving a critical gap in evaluating LLMs for civil law systems that govern most jurisdictions worldwide. To address this gap, we introduce Vietnamese Legal Benchmark VietLegal, a cognitively grounded benchmark designed for the hierarchical and codified structure of Vietnamese law. Although instantiated in Vietnamese legislation, VietLegal provides a replicable evaluation framework for civil law systems characterized by complex statutory hierarchies and frequent amendments. Inspired by Bloom’s taxonomy, VietLegal assesses multiple levels of legal understanding through tasks that mirror real-world legal assistant use cases, including legal question answering, multi-step reasoning, and scenario-based problem solving. The benchmark contains 10,450 expert-annotated samples, each cross-validated against authoritative legal sources to ensure fidelity to practical legal workflows. By offering the first standardized legal benchmark for Vietnamese, VietLegal enables systematic assessment of LLMs in civil law contexts and supports the development of more reliable and interpretable AI-assisted legal systems.

LLM benchmarks are often treated as coherent measurement units, yet they are heterogeneous collections of instances spanning diverse domains, skills, formats, and contexts. As a result, aggregate benchmark scores can conflate model capability with benchmark composition, obscuring what existing benchmarks actually cover. We introduce BenchHub, a composition-aware evaluation framework that represents benchmarks as distributions over instance-level attributes. BenchHub integrates 54 benchmarks comprising 839K samples across 10 languages. It enables researchers to inspect benchmark contents, uncover reusable coverage hidden behind benchmark names, transparently compare benchmarks, and construct controllable evaluation sets for application-aligned model selection. Using BenchHub, we show that similarly motivated benchmarks can differ substantially in internal composition, existing benchmarks often contain reusable coverage beyond their stated purposes, models exhibit fine-grained category-level performance variation hidden by aggregate scores, and model rankings can shift under different reweighting and resampling configurations. Our results motivate evaluation practices that make benchmark composition explicit, inspectable, and controllable.

Adaptive immune receptor repertoire sequencing data has emerged as a promising potential modality for disease diagnosis, relying on computational methods to analyze T-cell receptor (TCR) sequences from an individual’s blood sample. Published methods rely on different cohorts, data preprocessing pipelines, and evaluation metrics, making direct comparison across methods challenging. We have developed BenchRep-T, a unified benchmark that standardizes multiple publicly available TCR repertoire datasets and evaluates nine computational approaches, spanning statistical enrichment of shared sequences, feature-engineered ensembles, deep learning, and graph-based clustering. BenchRep-T evaluates methods on four tasks: disease classification across conditions, performance scaling under restricted sequence-sampling depth, recovery of known antigen-specific driver sequences, and evaluation of sensitivity to demographic confounding. Under controlled evaluation, simple baselines prove competitive, with tree-based models trained on V- and J-gene usage and short sequence motifs approaching the classification performance of more complex methods. Our findings underscore the complexity of modeling TCR repertoire data, and show that no single method dominates across all tasks. BenchRep-T provides a framework for rigorous and reproducible evaluation of TCR repertoire classification methods to accelerate the development of immune repertoire-based diagnostics.

We study fixed-confidence best arm identification in generalized linear bandits under a hybrid feedback model: at each round, the learner may query either (i) absolute reward feedback from a single arm or (ii) relative (dueling) feedback from an arm pair, both governed by generalized linear models. We introduce a likelihood-ratio–based confidence sequence that unifies heterogeneous generalized linear observations and yields an explicit ellipsoidal confidence set under a self-concordance assumption. Building on this confidence set, we propose a hybrid Track-and-Stop algorithm that adaptively allocates queries by tracking a minimax-optimal design over a joint action space of arms and pairs. We establish $\delta$-correctness and provide high-probability upper bounds on the stopping time. We further extend the framework to a cost-aware setting that accounts for heterogeneous acquisition costs across feedback modalities. Empirical experiments demonstrate that the proposed algorithms significantly improve sample efficiency over baseline methods.

Evaluating large language model (LLM) agents on scientific tasks typically reduces to final-task accuracy, which obscures where reasoning succeeds or breaks down. We introduce a benchmark and diagnostic evaluation framework for hypothesis-driven adaptive experimental planning, instantiated in kinetic mechanism identification. The framework goes beyond final accuracy in two ways. First, we evaluate exploration behavior by quantifying the extent to which agents exhibit space-filling or corner-seeking patterns in the experimental design space. Second, it decomposes the reasoning process into four interpretable steps; belief update, hypothesis discrimination, discriminative experiment design, and predictive reasoning under interventions, each scored independently on reasoning traces. Across ten LLM agents, we find that stronger agents outperform both adaptive (Bayesian optimization) and non-adaptive (Latin hypercube) baselines and exhibit exploration patterns inconsistent with naive heuristics. Decomposing reasoning reveals that the stronger agents score consistently well across all four steps, whereas weaker agents often succeed at belief update and hypothesis discrimination but fail at discriminative experiment design and predictive reasoning. This contrast shows that our framework identifies a specific reasoning bottleneck: weaker models can recognize plausible hypotheses and articulate their differences, but struggle to translate this understanding into concrete predictions and experimental designs. The code and dataset are available at https://github.com/krfdq48k4p-arch/HypothesisDrivenExperimentPlanningBench and https://huggingface.co/datasets/xq8wvm/HypothesisDrivenExperimentPlanningBench, respectively.

Comparing whether two dynamical systems implement the same computation despite differences in coordinates or measurements is a central problem in neuroscience and machine learning. Dynamical Similarity Analysis [DSA; Ostrow et al., 2023] addresses this problem by aligning finite-dimensional Koopman approximations of the two systems through an orthogonal similarity transformation. Here we show that orthogonal alignment is neither necessary nor sufficient for topological conjugacy: genuinely conjugate systems may be related by a non-orthogonal basis-transfer matrix that DSA cannot capture, while non-conjugate systems may have orthogonally equivalent Koopman operators that DSA will fail to distinguish. We then use this observation to formulate \emph{Conjugacy-based Similarity Analysis} (CSA), which restricts alignments to those induced by candidate state-space bijections rather than arbitrary orthogonal matrices. We prove that CSA's fitted alignment is the finite-data projection of the composition operator associated with the candidate bijection, and use controlled examples to show why this distinction matters when observable dictionaries are chosen explicitly or implicitly from data. Together, these results clarify what Koopman-based similarity measures must ensure to support claims of identifying conjugacies between computational systems.

Autoregressive graph generation is commonly formulated by linearizing a graph into a sequence via a flat walk over its nodes and edges. While effective for small molecular graphs, such representations do not scale gracefully to large, structurally complex biological graphs, where higher-order organization is both dense and functionally important. To address this limitation, we introduce a hierarchical graph abstraction together with a coarse-to-fine autoregressive pipeline that separates coarsening, tokenization, and decoding, and we study sequence constructions that expose hierarchy in the tokenization. Concretely, we compare a flat representation with two hierarchical alternatives: one that explicitly factorizes communities and their interactions, and another that induces hierarchy through a depth-first traversal. Across eight benchmarks, differences are modest on simpler domains but become pronounced on larger, more structurally complex settings such as cyclic peptides, proteins, and hard conditional-generation tasks. In these regimes, flat representations degrade in validity or distributional fidelity, while the traversal-based hierarchical representation is consistently more reliable; the explicitly factorized variant is generally weaker, suggesting that how hierarchy is exposed in the sequence matters. Finally, the same coarse-to-fine representation naturally enables motif-conditioned generation without architectural changes: on larger and more structurally complex graphs, conditioning on higher-level prefixes yields strong motif retention together with high validity, uniqueness, and novelty, whereas flat representations can exhibit severe mode collapse under comparable constraints.

It remains unclear whether quantum machine learning (QML) truly holds an advantage when tackling practical and meaningful tasks. In exploring this question, the importance of classical data encoding, often overlooked, becomes increasingly evident. Amplitude encoding, which can embed $2^n$ classical data into $n$ qubits, is widely used due to its apparent efficiency. However, its potential limitations for QML have yet to be fully explored. In this paper, we establish a theoretical result for the limitations caused by quantum encoding and point out the existence of a concentration phenomenon in amplitude encoding. Compared to prior work, our theoretical result operates under more general conditions and leads to a stronger conclusion. This concentration phenomenon causes the classification predictions to remain close to random guessing, regardless of the training process. Our findings shed new light on a long-standing puzzle in the field of QML: why some QML models perform well on simple datasets like MNIST but fail to generalize to more complex practical tasks. By highlighting the pivotal role of encoding design in QML, our work clearly indicates that future research needs to focus more on the design of classical data encoding to advance the effectiveness of QML.

Although diffusion models achieve promising image generation performance, uncertainty during the iterative denoising process can lead to visual artifacts. Existing uncertainty estimation methods typically quantify it at the pixel level. However, these methods assume independence among pixels, neglecting pixel correlations crucial for image structure. In contrast, we propose a frequency-domain uncertainty estimation method that captures structural correlations. We empirically show that samples with artifacts exhibit higher frequency-domain uncertainty, and derive a Louis-identity-based connection between the estimated uncertainty and the optimal reverse-process covariance. To this end, we develop a structure-aware diffusion sampling guidance framework. For the mean of the reverse process, the gradient of the uncertainty is used to penalize specific frequency components; for the reverse covariance, frequency-domain uncertainty is used as a proxy to modulate injected noise. Experiments on the ImageNet, LSUN-Churches, and DrawBench datasets across U-Net, U-ViT, and SD3 architectures validate our method.

Standard Retrieval Augmented Generation (RAG) is poorly matched to agent memory. Unlike large heterogeneous corpora, agent memory forms a bounded and coherent interaction stream in which many spans are highly correlated or near duplicates. As a result, flat top-$k$ similarity retrieval often returns redundant context, while summary-centric hierarchies can blur the subtle details that distinguish one candidate from another. We argue that agent memory should follow the principle of decoupling before aggregation: the system should first isolate reusable facts, updates, and distinguishing details from similar histories, and only then organise them for efficient retrieval. Based on this principle, we propose xMemory, which constructs a revisable hierarchical memory structure from original messages to segments, memory components, and groups. xMemory segments interaction history into local events, decouples each segment into memory components, aggregates related components into high-level groups using a sparsity--semantic faithfulness objective, and maintains this structure incrementally as memory evolves. At inference time, xMemory retrieves top-down, first selecting a compact backbone of complementary groups and components, and then expanding to segments and raw messages only when additional evidence reduces the reader's uncertainty. Experiments on LoCoMo and PerLTQA across diverse open source and closed source LLMs show consistent gains in answer quality and inference token efficiency, supported by analyses of redundancy, evidence density, and coverage.

Visual-token compression is effective for improving the efficiency of vision-language models, but under extreme compression budgets, token pruning can break visual grounding while learned resamplers increase parameter count, attention cost, and training complexity. We revisit compression through a token parameterization lens, separating (i) basis transformation and structured truncation (retained subspace/compressibility) from (ii) coordinate organization (optimization and cross-modal alignment). This view yields two coupled objectives, compressibility and learnability, which we formalize as unified functionals. Guided by these objectives, we design Braco, a lightweight four-step coder that combines transform-basis truncation, input-independent basis-coordinate embeddings, budget-dependent orthogonal re-parameterization, and learned spatial residual tokens from lightweight pooling. Experiments show that Braco forms the favorable empirical accuracy-efficiency frontier under compression ratios from $23\times$ to $64\times$ and remains competitive at $144\times$, reaching 95.2\% accuracy while reducing prefill FLOPs by 84.2\%--86.7\% relative to the uncompressed upper bound. Against prior methods, Braco matches or improves accuracy while achieving up to $\sim$36\% end-to-end speedup and using $16.6\times$/$78.8\times$ lower compressor latency/FLOPs.

Multimodal Large Language Models (MLLMs) have demonstrated substantial promise in spatial understanding. Existing works typically incorporate prior knowledge extracted from a pre-trained foundation model to further enhance the spatial awareness of MLLMs. In this paper, we first reveal that when integrating diverse foundation models into MLLMs, each model provides complementary spatial priors that benefit different tasks. Motivated by this, we propose $\textbf{ViPS}$, a novel multi-model prior framework designed to fully unleash the potential of incorporating multiple $\textbf{Vi}$sual $\textbf{P}$riors from diverse models into MLLMs for $\textbf{S}$patial understanding. Specifically, ViPS introduces an Efficient Prior Proxy to generate multiple foundational priors with minimal inference overhead, and a Dynamic Prior Fusion mechanism to achieve harmonious and context-aware prior fusion and injection from the prior proxies. Extensive experiments demonstrate that ViPS successfully harmonizes diverse visual priors, establishing new state-of-the-art performance across multiple complex spatial reasoning and 3D spatial understanding benchmarks.

Contrastive vision-language (VL) pretraining in computational pathology has generally relied on curated image-caption datasets sourced from social media, educational videos, and research publications. These corpora are sparse relative to the diversity of tissue morphology and difficult to scale further, as pathologists do not routinely produce descriptive captions in clinical practice. We propose to instead leverage slide-level pathology reports which can be generated during routine diagnostic workflows in a scalable manner. A key challenge in this weakly-supervised setting is that pathology reports describe macroscopic, diagnostic findings while patch-level visual features describe microscopic findings. We introduce Locked Image Multiple Instance PreTraining (LIMIT), which addresses this gap by freezing a pretrained vision encoder and fine-tuning only the text encoder: report embeddings serve as cross-attention queries over bags of patch features, producing report-conditioned slide representations optimized via cross-entropy. Using LIMIT, we establish CONCH-Z, which aligns the text encoder of CONCH v1.5 to 335,645 clinical reports. Evaluated across 20 tasks spanning patch-level classification, tumor detection, and slide-level subtyping, CONCH-Z establishes state-of-the-art zero-shot performance over VL encoders trained on curated caption datasets, while simultaneously closing the gap with slide-level foundation models trained with substantially more parameters and compute. We will release CONCH-Z weights and evaluation code to support reproducibility and broader community use.

Activation steering has emerged as a promising alternative for controlling language-model behavior at inference time by modifying intermediate representations while keeping model parameters frozen. However, large-scale evaluations such as AxBench show that existing steering methods are often outperformed by simple in-context prompting and generalize poorly to unseen concepts. We hypothesize that these limitations arise from unvalidated simplifying assumptions shared across prior methods, which typically restrict steering interventions to fixed, single-step, position-invariant transforms. We propose FLAS (Flow-based Activation Steering), which learns a general, concept-conditioned velocity field $v_t(h,t,c)$ that transports unsteered activations to steered ones without relying on these assumptions. On AxBench, FLAS is the first learned method to consistently outperform prompting, reaching held-out harmonic means of $1.015$ on Gemma-2-2B-IT and $1.113$ on Gemma-2-9B-IT without per-concept tuning. Analysis of the learned flow shows curved, multi-step, token-varying trajectories, which suggests that previous hypotheses on activation space geometry might be incomplete. Our code is available at https://anonymous.4open.science/r/FLAS.

Vision-Language-Action (VLA) models have advanced rapidly with stronger backbones, broader pre-training, and larger demonstration datasets, yet their action heads remain largely homogeneous: most directly predict action commands in a fixed world coordinate frame. We propose \textbf{MCF-Proto}, a lightweight action head that equips VLA policies with a Motion-Centric Action Frame (MCF) and a prototype-based action parameterization. At each step, the policy predicts a rotation $R_t \in SO(3)$, composes actions in the transformed local frame from a set of prototypes, and maps them back to the world frame for end-to-end training, using only standard demonstrations without auxiliary supervision. This simple design induces stable emergent structure. Without explicit directional labels, the learned local frames develop a stable geometric structure whose axes are strongly compatible with demonstrated end-effector motion. Meanwhile, actions in the learned representation become substantially more compact, with variation captured by fewer dominant directions and more regularly organized by shared prototypes. These structural properties translate into improved robustness, especially under geometric perturbations. Our results suggest that adding lightweight geometric and compositional structure to the action head can materially improve how VLA policies organize and generalize robotic manipulation behavior. An anonymized code repository is provided in the supplementary material.

Large Reasoning Models (LRMs) have shown excellent performance in reasoning tasks using long Chain of Though. However, this has also led to a significant increase of computational costs and the generation of verbose output, a phenomenon known as overthinking. The tendency to overthinking is often exacerbated by Reinforcement Learning (RL) algorithms such as GRPO/DAPO. In this paper, we propose BFS-PO, an RL algorithm which alleviates this problem using a Best-First Search exploration strategy. Specifically, the reasoning chains generated at training time by BFS-PO are organized into a search tree, within which the shortest correct sequence is selected as the best and expanded using a backtracking mechanism based on maximum entropy nodes. In this way, we bias the exploration of the solution space towards the search for increasingly shorter solutions, training the LRM to generate progressively more concise answers. Using different benchmarks and base LRMs, we show that BFS-PO can simultaneously increase the LRM accuracy and shorten its reasoning chains. Our code and models are available in the supplementary material and will be published after this article is accepted.

While LLMs excel at single-turn generation, they struggle with long-horizon, multi-turn interactions. Offline reinforcement learning (RL) offers a scalable approach, yet its performance hinges on the availability and quality of multi-turn trajectory data. A common remedy is to augment training with synthetic trajectories generated by LLMs or simulators, but synthetic data is highly heterogeneous in quality, and naively treating all trajectories as equally informative can degrade performance. We propose BOOST, a bilevel optimization framework where the inner level trains the LLM on reweighted data and the outer level trains a lightweight reweighting head on held-out real validation tasks, assigning continuous trajectory-level weights without requiring an external judge. To ground this approach, we derive a PAC-Bayesian bound revealing a three-way trade-off: synthetic data increases diversity but risks task-shift, while concentrating weight on high-quality trajectories improves empirical performance at the cost of effective sample size. Empirically, our method consistently outperforms multiple baselines. Analysis reveals it upweights synthetic trajectories that align with the real data distribution and exhibit higher qualitative merit.

In large-scale outdoor LiDAR localization, Scene Coordinate Regression (SCR) achieves sub-meter accuracy, but its deployment in high-precision autonomous driving is hindered by a "last-mile" problem: the point-wise prediction paradigm causes unphysical trajectory jittering, rendering localization kinematically discontinuous despite low mean errors. To bridge the gap between high-precision localization and temporal kinematic consistency, we propose BiLi, a spatiotemporal synergistic relocalization framework. Spatially, BiLi utilizes Implicit Manifold Regularization (ISR) driven by an offline neural implicit field. Formulated as a novel point-to-manifold projection constraint, ISR strictly anchors predicted coordinates to continuous physical zero-level sets. Temporally, to bypass computationally heavy implicit optimization, we distill the teacher's kinematics into a lightweight feed-forward network, compelling it to learn robust, equivariant physical priors. Finally, a Differentiable Robust Manifold Gating mechanism dynamically fuses the global spatial predictions with the temporal kinematic states via adaptive state updating, facilitating efficient real-time deployment. Extensive experiments demonstrate that BiLi fundamentally resolves the "last-mile" challenge. Our method outperforms state-of-the-art approaches by 25% and 65% on the Oxford and NCLT benchmarks, respectively, delivering drift-free, kinematically coherent, and highly accurate real-time localization.

We study \emph{bilinear matching bandits}, an online sequential decision problem in which at each round a learner assigns $M$ users to $N$ items via an injective matching $\pi:[M]\hookrightarrow[N]$ and observes semi-bandit feedback with bilinear mean reward $\mathbf{x}_i^\top \mathbf{\Theta}^\star \mathbf{y}_{\pi(i)}$, where the parameter $\mathbf{\Theta}^\star \in \mathbb{R}^{d_1 \times d_2}$ is unknown. We propose \textsc{HybridRRD}, a hybrid algorithm that combines high-width elimination with a row-wise regularized design sampler over the fractional matching polytope. Without assuming $\mathbf{\Theta}^\star$ is low-rank, \textsc{HybridRRD} achieves a high-probability regret bound of $\widetilde{O} (d_2 \sqrt{d_1MT} + Md_1 d_2)$, suppressing problem-scale constants. This improves the leading dimension dependence from $d_1d_2$ to $d_2\sqrt{d_1}$ over the standard Kronecker-linearized contextual combinatorial semi-bandit baseline. We further prove a minimax lower bound of $\Omega(d_2 \sqrt{d_1 M T})$ in a large-item regime, matching the leading regret term up to logarithmic factors. Numerical experiments show that \textsc{HybridRRD} performs favorably against Kronecker-linearized baselines.

Reliable use of large language models (LLMs) requires uncertainty estimates that separate ambiguity intrinsic to a prompt from genuine knowledge gaps. Bayesian and deep-ensemble methods deliver this aleatoric/epistemic decomposition in principle, but are computationally prohibitive at LLM scale and require white-box access. Existing black-box methods, in contrast, summarize sample variability into a single consistency scalar that cannot make the distinction. We introduce a two-level ensemble that bridges this gap. Stochastic decoding of a fixed prompt forms the inner ensemble; meaning-preserving semantic perturbations of the prompt form the outer ensemble. Variability is measured in a continuous embedding space, and the law of total covariance gives an exact AU/EU split that requires only black-box sampling access. On top of this estimator, we contribute three theoretical results that ground the framework: (i) a finite-sample bias identity for nested Monte Carlo with a closed-form bias-corrected epistemic estimator; (ii) a Bayesian bridge bounding the gap between perturbation-population AU/EU and Bayesian targets through three local conditions that we estimate as diagnostics from the data the estimator already collects; and (iii) a separation result identifying a class of failure modes that any consistency-only black-box method provably misses while perturbation-based EU detects. Across five short-form QA benchmarks and four instruction-tuned LLMs spanning two families and two scales, the estimator matches or surpasses strong white-box baselines; controlled paired interventions on AmbigQA and Natural Questions confirm the AU/EU split tracks distinct, actionable sources of failure; and a long-form pilot on the NQ long-answer split shows the framework extends to claim-level factuality without retraining. Across five short-form QA benchmarks and four instruction-tuned LLMs spanning two families and two scales, the estimator matches or surpasses strong white-box baselines; controlled paired interventions on AmbigQA and Natural Questions confirm the AU/EU split tracks distinct, actionable sources of failure; and a long-form pilot on the NQ long-answer split shows the framework extends to claim-level factuality without retraining.

Multimodal Large Language Models (MLLMs) have shown promise as interactive agents, yet precise, programmatic 3D graphics editing remains difficult to train because existing systems largely rely on closed-source model APIs and lack deterministic, state-verifiable execution environments. We present BlenderFORGE, a trainable data-environment-optimization framework for open-weight MLLM agents that edit explicit Blender scenes through Blender Python (bpy) code generation. BlenderFORGE targets four fundamental state-verifiable editing task families: object placement, shape-key editing, lighting adjustment, and material modification. It integrates three components: (1) a scalable perturbation-based data generation pipeline built from curated assets and large-scale 3D scene repositories; (2) an isolated Blender Sandbox for deterministic code execution, structured state export, and closed-loop reward computation; and (3) a two-stage training pipeline that uses offline teacher trajectories for cold-start Supervised Fine-Tuning (SFT) and sandbox-computed rewards for multi-turn tool-agentic Reinforcement Learning (AgentLoop-RL). Experiments across the four task families show that the resulting open-weight Qwen3-VL-8B-BlenderFORGE agent substantially improves over its base model and achieves competitive performance with several zero-shot proprietary MLLM baselines.

We study long-context balanced entropic optimal transport (OT) attention on TPU hardware through a stopped-base, fixed-depth tail-refinement surrogate. After a stopped $T$-step Sinkhorn solve, we unroll a short refinement tail and differentiate that surrogate exactly. For the production $R=2$ case, the backward pass contains four staircase plan factors. We prove an exact one-reference-tile schedule: the $R=2$ score cotangent is a single reference plan tile times an explicit modifier field built from vector cotangents and dual differences. This yields block-wise cost $O((T+R)LW)$, $O(Ld)$ input storage, and $O(L)$ additional HBM usage for fixed head dimension $d$ and band width $W$. We also formalize the current \texttt{dustbin\_block} path as the same balanced surrogate on an augmented support, so the schedule lifts to the gap-aware transport path used in our TPU runs. We provide a local surrogate-bias bound, an a posteriori bias certificate, and a projective contraction certificate for strictly positive active blocks. On synthetic masked problems, the optimized kernel matches exact autodiff of the same centered surrogate to within $10^{-5}$--$10^{-10}$. On TPU v6e-8, a four-configuration Pfam screen completes end-to-end, and a promoted balanced $R=2$ run sustains roughly $8.5$ examples per second through a three-hour budget, reaching step $1437$. Held-out Pfam test shards improve reconstruction from $3.17$ to $0.99$ and sparse CE from $5.86$ to $5.69$ relative to step $0$. These results support exact fixed-depth backward theory, a theorem-matching gap-aware bridge, and trainability evidence for the production path.

The deployment of Large Language Models (LLMs) with extended context windows is fundamentally constrained by the quadratic computational and memory costs of the self-attention mechanism. While acceleration techniques like sparse attention and quantization offer potential relief, they currently face a critical dilemma: uniform compression strategies fail to capture the non-uniform distribution of information importance, degrading performance on long-range dependencies; conversely, fine-grained, token-level adaptation often introduces irregular memory access patterns that negate theoretical efficiency gains on modern GPUs. To resolve this tension, we introduce BMAttn (Block-Aligned Mixed-Precision Attention), a framework that unifies importance-aware precision allocation with hardware-friendly execution. BMAttn partitions attention maps into block-aligned high-precision, low-precision, and sparse regions, using a novel affine windowing mechanism to dynamically adjust boundaries based on sequence length. We further propose a saliency-weighted calibration method and a layer-adaptive regularizer that adaptively aligns precision targets with perceptual importance and layer sensitivity. Extensive experiments demonstrate that BMAttn achieves a \textbf{3.3x} speedup on long-context tasks with negligible accuracy loss, and up to \textbf{5x} speedup with minimal degradation, effectively bridging the gap between algorithmic adaptivity and hardware efficiency.

Adversarial perturbation defenses protect images from generative editing by adding imperceptible noise. Yet none of these methods have been evaluated against AI nudification, despite AI-generated non-consensual intimate imagery (AIG-NCII) being among the most documented harms of generative AI and produced largely through inpainting. We argue this is a problem-formulation failure: prior threat models do not correspond to nudification attacks, and the misalignment is structural, spanning mask shape, prompt distribution, and success criterion which causes existing evaluations to systematically overestimate protection effectiveness. We reformulate adversarial inpainting protection around this documented harm. We introduce BodyBench, a benchmark of 648 fully synthetic clothed adult subjects, each paired with 11 mask variants spanning body, face, context, and dilation regions, along with nudification and redressing attack prompts. We propose three harm-aligned metrics: sexualization, naturalness, and recognizability that jointly determine whether an inpainted output constitutes a successful nudification attack. Auditing four recent protection methods (PhotoGuard, DiffusionGuard, AdvPaint, DiffVax) on BodyBench, we find that protection succeeds in only one configuration of one method, and only under exact mask alignment between defender and attacker---a condition no real defender can guarantee and one that we show is easily always bypassed. We also show that standard metrics such as PSNR are essentially uncorrelated with nudification protection. BodyBench makes this misalignment measurable and provides a foundation for defenses grounded in documented harm.

Reinforcement Learning with Verifiable Rewards (RLVR) has become an important approach for improving the reasoning capabilities of large language models (LLMs). On-policy training remains the dominant RLVR paradigm, but its reliance on fresh rollouts leads to poor data efficiency. To address this bottleneck, off-policy paradigm has attracted increasing attention by reusing historical trajectories through replay buffers. Despite its efficiency advantages, off-policy training often struggles to match the performance of on-policy training. Our analysis reveals that this limitation primarily stems from data staleness introduced by replay buffers: to maintain training stability, mainstream algorithms discard a large fraction of high-staleness optimization signals, leading to substantial performance degradation. To address this issue, we propose STAR (Staleness-Aware Replay), a data-centric replay framework decoupled from specific optimization algorithms. STAR improves off-policy RLVR by constructing low-staleness and high-quality training data. Experiments on a range of mathematical reasoning and code generation tasks show that STAR consistently improves off-policy RLVR with up to 19% relative performance gains, and enables it to surpass the on-policy baseline using only around 13% of the rollout data while achieving 5x-6x wall-clock speedup.

Code agents are rapidly lowering the barrier for developers to work with unfamiliar repositories, and are increasingly adopted for tasks such as building prototypes, reproducing experiments, and adding features. Reliably executing these tasks depends on a critical prerequisite: the target repository must first be successfully set up and brought into a usable development state. This bootstrapping process requires substantial trial-and-error exploration by code agents, yet the resulting knowledge (e.g., resolved dependencies, repair strategies) remains trapped in a single conversation, unavailable to future agents or developers. We therefore formulate repository bootstrapping as a reusable startup knowledge problem and introduce BootstrapAgent, a multi-agent framework that distills bootstrap exploration into a persistent, verifiable, agent-consumable .bootstrap contract. Through CI evidence extraction, structured planning, deterministic Docker-based verification, and trace-driven repair, BootstrapAgent generates a contract covering environment setup, diagnostic checks, minimal verification, and accumulated repair knowledge. We further propose warm repair with clean replay to accelerate iterative debugging without sacrificing cold-start reproducibility, and a two-level verification strategy to prevent reward hacking. Experiments on three benchmarks show that BootstrapAgent achieves a 92.9% success rate, outperforming the baseline by over 10% while reducing downstream agent token usage by 25.9% and build time by 22.3%.

Brain foundation models have achieved remarkable advances across a wide range of neuroscience tasks. However, most existing models are limited to a single functional modality, restricting their ability to exploit complementary spatiotemporal dynamics and the collective data scale across different neuroimaging techniques. This limitation largely arises from severe semantic heterogeneity and resolution discrepancies among modalities. To address these challenges, we propose Brain-OF, an omnifunctional brain foundation model jointly pretrained on fMRI, EEG and MEG, capable of handling both unimodal and multimodal inputs within a unified framework. To reconcile heterogeneous spatiotemporal resolutions, we introduce the Any-Resolution Neural Signal Sampler, which projects diverse brain signals into a shared semantic space. To further manage semantic shifts, the Brain-OF backbone integrates DINT attention with a Sparse Mixture of Experts, where shared experts capture modality-invariant representations and routed experts specialize in modality-specific semantics. Furthermore, to explicitly internalize the characteristics of neural activity through self-supervised learning, we propose Masked Temporal-Frequency Modeling, a dual-domain pretraining objective that jointly reconstructs brain signals in both the time and frequency domains. Brain-OF is pretrained on a large-scale corpus comprising around 40 datasets and demonstrates superior performance across diverse downstream tasks, highlighting the benefits of joint multimodal integration and dual-domain pretraining. Our code will be made publicly available upon acceptance.

Token-level dynamic sparse attention exemplified by DeepSeek Sparse Attention (DSA) selects the globally most relevant key-value tokens via an exact Top-$K$ operator, achieving superior model quality over block-level alternatives. However, this exact selection creates a severe distributed inference bottleneck: enforcing an exact global Top-$K$ across GPUs inevitably incurs either redundant full-context retrieval or costly multi-stage cross-device synchronization, which largely negates the computational advantages of DSA at long context lengths. We first show that the exact Top-$K$ bound is unnecessary during inference: once the truly critical tokens are recalled, admitting additional context preserves or even improves accuracy. Leveraging this insight, we propose Interleaved DeepSeek Sparse Attention (IDSA), which distributes tokens across GPUs in an interleaved layout so that each device performs only a relaxed local Top-$m$ selection. Under this layout, the union of independent per-GPU Top-$m$ selections near-completely covers the globally most relevant Top-$K$ tokens. This allows each device to proceed with its local selection with minimal cross-GPU overhead while avoiding both expensive full-context Top-$K$ computation and multi-stage cross-GPU merging, enabling a not only distributed but also synchronization-efficient inference pipeline. Without any retraining, IDSA delivers dramatic throughput gains for context lengths exceeding 100K tokens on both DeepSeek-V3.2 and GLM-5, while preserving equivalent or better reasoning performance on the AIME and Needle-In-A-Haystack benchmarks.

Tabular data synthesis aims to generate high-quality data while preserving privacy. However, we find that existing tabular generative models exhibit a clear tradeoff in the small-data regime: improving data quality typically comes at the cost of increased memorization of training samples, thereby weakening privacy protection. This tradeoff arises because small training sets make it difficult for dataset-specific generative models to distinguish generalizable structure from sample-specific patterns. To address this, we propose DiffICL, which formulates tabular data generation as an in-context learning problem. Instead of fitting each dataset from scratch, DiffICL leverages pretrained structural priors learned from a large collection of datasets, enabling it to infer data distributions from limited context rather than memorizing individual samples. We evaluate DiffICL on 14 real-world datasets. Results show that DiffICL improves both data quality and privacy, and generates synthetic data that provides effective data augmentation. Our findings suggest that the quality–privacy tradeoff can be improved through better training paradigms.

Decoding visual content from non-invasive brain signals remains challenging because neural responses evolve over time whereas images are typically static. Most existing methods align an entire neural response window to a single final visual embedding, this overlook a fundamental representational mismatch: brain signals reflect both early perceptual and later integrative representations, whereas the final visual embedding is biased toward high-level semantics. Therefore, we propose BRIDGE, a brain-vision representation integration framework that aligns two modalities through Visual Depth Encoding and Brain Granularity Encoding. On the visual side, BRIDGE extracts and fuses CLIP representations from multiple depths, producing an alignment target that preserves low-level and high-level information. On the brain side, instead of treating the whole temporal window as homogeneous and obscuring temporally heterogeneity, BRIDGE explicitly partitions stimulus-evoked EEG/MEG responses into a small number of temporally ordered stages and adaptively pools them. The resulting brain and visual embeddings are trained in a shared latent space by contrastive learning and can further support brain-to-image generation through a pretrained diffusion prior. Experiments on THINGS-EEG and THINGS-MEG demonstrate that BRIDGE achieves strong retrieval and generation performance. Ablation studies further confirm the complementary benefits of depth-wise visual aggregation and temporal brain factorization.

Attributed Graph Clustering (AGC) is a fundamental unsupervised task that integrates structural topology and node attributes to uncover latent patterns in graph-structured data. Despite its significance in industrial applications such as fraud detection and user segmentation, a significant chasm persists between academic research and real-world deployment. Current evaluation protocols suffer from the small-scale, high-homophily citation datasets, non-scalable full-batch training paradigms, and a reliance on supervised metrics that fail to reflect performance in label-scarce environments. To bridge these gaps, we present PyAGC, a comprehensive, production-ready benchmark and library designed to stress-test AGC methods across diverse scales and structural properties. We unify existing methodologies into a modular Encode-Cluster-Optimize framework and, for the first time, provide memory-efficient, mini-batch implementations for a wide array of state-of-the-art AGC algorithms. Our benchmark curates 12 diverse datasets, ranging from $2.7 \times 10^3$ to $1.1 \times 10^8$ nodes, specifically incorporating industrial graphs with complex tabular features and low homophily. Furthermore, we advocate for a holistic evaluation protocol that mandates unsupervised structural metrics and efficiency profiling alongside traditional supervised metrics. Our benchmark offers the community a robust, reproducible, and scalable platform to advance AGC research towards realistic deployment.

Drug-target interaction (DTI) prediction is pivotal for accelerating drug discovery, yet existing methods struggle with generalization under cross-domain distribution shifts. Conventional approaches typically rely on static protein representations and assume distributional consistency between source and target domains, often resulting in poor calibration when encountering novel protein families or scaffolds. To address this issue, we present $\textbf{Bi}$-view $\textbf{Fold}$-aware prediction (BiFold), a novel unified interaction-adaptive framework that enhances generalization via dual-view representation learning and calibration-aware adaptation. BiFold synergistically models sequence and structural evidence, introducing a routing module to dynamically fuse the two views at the feature-channel level and exploit complementary information. Furthermore, we design a calibration-aware training principle that aligns domain statistics and encourages flatter source solutions. Crucially, target supervision is introduced only after a forward-only warm-up phase to prevent reinforcement of miscalibrated early predictions, effectively mitigating confirmation bias in pseudo-label learning. Extensive experiments demonstrate that BiFold consistently outperforms the state-of-the-art methods across diverse datasets in both in-domain and cross-domain settings. Additionally, interpretability analysis reveals that the routing module partitions the feature space into view-dominant and view-neutral subspaces, providing concrete evidence of adaptive multi-modal complementarity.

Many existing defenses mainly strengthen refusal behavior to improve the safety of LLMs. Recent interpretability work shows that, in safety-aligned LLMs, harmfulness belief and refusal behavior are represented separately in the hidden states at the last token of the user instruction ($t_{\mathrm{inst}}$) and the response-start token ($t_{\mathrm{post}}$), respectively. Based on it, we find that harmfulness belief is not reliably preserved when it is carried from the $t_{\mathrm{inst}}$ to the $t_{\mathrm{post}}$. As a result, even when the LLM internally recognizes a request as harmful, it may still produce a compliant response. To address this, we propose Bridging Harmfulness and Refusal (BHR), a training framework that bridges harmfulness belief and refusal behavior. Specifically, BHR trains the adapter so that harmfulness information is carried to the hidden state at the $t_{\mathrm{post}}$, thereby establishing a second pathway from harmfulness belief to refusal behavior. To keep this signal reliable during fine-tuning, we introduce two additional losses. A belief loss helps the LLM maintain an accurate harmfulness judgment at the $t_{\mathrm{inst}}$. A belief-gated refusal loss uses the LLM's harmfulness belief at $t_{\mathrm{inst}}$ to regulate refusal learning, making attacks that target refusal features less effective while reducing the risk of over-refusal on benign inputs. Experiments across multiple LLMs and safety benchmarks show that BHR substantially improves robustness against both white-box and black-box jailbreak attacks, reduces the risk of over-refusal on benign prompts, and preserves general capabilities.

Real-world time series typically exhibit non-uniform information density across temporal positions, frequency structures, and forecasting horizons: critical dynamics are concentrated in local segments, different frequency components carry different structural information, and different horizons require different modeling complexity. However, most existing methods still allocate computation approximately uniformly across these dimensions, leading to redundant computation in low-information regions and insufficient modeling of key patterns. To address this, we propose BucpTSF, an information-density-driven structured framework for time series forecasting. Specifically, BucpTSF reorganizes raw sequences into hierarchically heterogeneous multi-level token representations through information-density-aware temporal representation reorganization, captures key cross-token dependencies with low overhead via frequency-selective structured token interaction, and further introduces horizon-conditioned structured extrapolation to adaptively balance dynamic modeling and structural extrapolation bias for different horizons, thereby improving the accuracy and stability of long-term forecasting. Extensive experiments on real-world datasets show that BucpTSF consistently outperforms strong baselines under various forecasting settings, validating the effectiveness of information-density-driven non-uniform computation allocation for time series forecasting.

Gradient clipping is a central but brittle design choice in differentially private federated learning: overly small thresholds bias client updates, whereas overly large thresholds amplify the noise required by privacy. This trade-off is further complicated by client heterogeneity and personalized privacy budgets, where a single global threshold is often mismatched and online adaptation from private training statistics can complicate privacy accounting. We study clipping-threshold selection as a privacy-budget-conditioned policy learning problem. We propose PAC-DP, a proxy-learned clipping framework for record-level locally private federated learning. Before private training, PAC-DP calibrates a deterministic policy $\pi_\theta(\varepsilon,t)$ from public or synthetic proxy simulations, mapping a client's target privacy budget and the training round to a clipping threshold. The policy is then frozen and used during private training, so deployed thresholds depend only on the declared budget, round index, and public/proxy-learned parameters, not on private gradients, losses, or client-specific update histories. PAC-DP combines this policy with per-example clipping, Gaussian perturbation, and per-client RDP accounting. We show that, under public or separately privatized proxy calibration, the frozen policy introduces no additional record-level privacy loss beyond the clipped Gaussian mechanisms. We further analyze the clipping-sensitive utility trade-off through a decomposition involving clipping bias, stochastic variance, client heterogeneity, and DP noise, and characterize proxy-to-target transfer via a regret bound. Experiments on MNIST, CIFAR-10, CIFAR-100, and Heart Disease show that PAC-DP improves the privacy--utility trade-off and communication efficiency over implemented fixed-threshold and adaptive DP-FL baselines under matched privacy budgets.

Understanding the reasoning behind deep learning model predictions is crucial in cheminformatics and drug discovery, where molecular design determines their properties. However, current evaluation frameworks for Explainable AI (XAI) in this domain often rely on artificial datasets or simplified tasks, employing data-derived metrics that fail to capture the complexity of real-world scenarios and lack a direct link to explanation faithfulness. To address this, we introduce B-XAIC, a novel benchmark constructed from real-world molecular data and diverse tasks with known ground-truth rationales for assigned labels. Through a comprehensive evaluation using B-XAIC, we reveal limitations of existing XAI methods for Graph Neural Networks (GNNs) in the molecular domain. This benchmark provides a valuable resource for gaining deeper insights into the faithfulness of XAI, facilitating the development of more reliable and interpretable models.

We propose a label-free post-processing framework that improves a strong but miscalibrated primary model using a weaker yet better-calibrated reference. Our key insight is that strict improvement is possible if and only if the two models are not *mutually calibrated*, meaning there does not exist joint distribution over their predictions and outcomes such that both predictions are simultaneously calibrated. We formalize this condition and connect it to no-arbitrage results from economics. Under such condition, we develop an efficient post-processing algorithm of the strong model's outputs based on Bregman projection, with a strict worst-case performance improvement guarantee. Experiments on representative LLMs across varying scales demonstrate the effectiveness of our method, reducing the ECE of the primary model by over $40$\% on common benchmarks.

Structure prediction models are increasingly deployed as zero-shot digital screens in antibody design, operating under the assumption that high folding confidence implies true binding. However, rigorous validation of this premise has been limited by small test sets, narrow target diversity, and the risk of data leakage from training-adjacent structures. Here we introduce POISK: Patent-extracted Organization of Immunoglobulin Sequence Knowledge, a large-scale dataset derived from the US patent literature, comprising 89074 antibody-antigen records extracted from 10110 patents using an LLM-based pipeline. From this resource, we curate a balanced benchmark set of 22494 complexes: 11247 patent-validated binders without a known resolved structure paired with putative non-binders constructed from sequence-similar human repertoire antibodies. Using this dataset, we benchmark six leading structure prediction models (AlphaFold-Multimer, Boltz-2, Intellifold v2, OpenFold3, Protenix-v1, and Protenix-v2) on their ability to discriminate binders from decoys. The best-performing model, Protenix-v2, achieves 87% pairwise accuracy on the most confident half of predictions and correctly identifies the true binder from a pool of 50 candidates in 32% of trials; this represents a substantial enrichment over random selection, yet remains far from reliable single-candidate identification. Structural consensus across models also discriminates binders from decoys, with pairwise accuracy reaching 73.1% when three or more models converge on a pose DockQ $\geq$ 0.23, covering around 54.2% of pairs). Yet epitope localization remains poor: the best model contacts annotated epitope residues in only 45% of cases, indicating that high confidence reflects learned statistical associations rather than accurate physical binding modes. Our results suggest that current folding models provide useful but limited signal for antibody screening and should be complemented by orthogonal methods. We release POISK as a community resource for future benchmarking and model training.

Most Vision-Language Models (VLMs) are built by extending pretrained Large Language Models (LLMs) with visual modules and multimodal alignment. However, this multimodal scaling often degrades the language-side reasoning ability originally encoded in the base LLM. While the base LLM retains usable reasoning ability after scaling, the aligned VLM itself cannot reliably access this ability. Therefore, recovering degraded reasoning capability in VLMs may be more effective when using the base LLM as the source rather than relying on the VLM alone. Motivated by this, we propose LIFT (Language-side reasoning Facilitation and Transfer), a lightweight vector-intervention method that transfers reasoning capability from the base LLM to the VLM without retraining the backbone. LIFT defines Reasoning Vectors as answer-token hidden-state differences between a Reasoner path with an explicit reasoning trace and a Solver path without it, and injects these vectors into language-side activations of the target VLM. LIFT further supports learnable vector adaptation while keeping the VLM backbone frozen. We evaluate LIFT on two VLMs across six reasoning benchmarks, comparing Reasoning Vectors extracted from the base LLM and from the aligned VLM under matched protocols. Results show that LLM-derived vectors consistently outperform VLM-derived vectors, confirming that the base LLM is a more effective source for recovering reasoning. LIFT partially recovers degraded reasoning through lightweight language-side interventions. Further analyses show that Reasoning Vectors influence intermediate reasoning behavior rather than merely altering final answers.

Operational disaster response goes beyond damage assessment, requiring responders to integrate multi-sensor signals, reason over road networks, populations and key facilities, plan evacuations, and produce actionable reports. However, prior work largely isolates remote-sensing perception or evaluates generic tool use, leaving the end-to-end workflows of emergency operations underexplored. In this paper, we introduce Disaster Operational Response Agent benchmark (DORA), the first agentic benchmark for end-to-end disaster response: 515 expert-authored tasks across 45 real-world disaster events spanning 10 types, paired with expert-verified, replayable gold trajectories totaling 3,500 tool-call steps. Tasks span five dimensions that cover the operational disaster-response pipeline: disaster perception, spatial relational analysis, rescue and evacuation planning, temporal evolution reasoning, and multi-modal report synthesis. Agents compose calls from a 108-tool MCP library over heterogeneous geospatial data: optical, SAR, and multi-spectral imagery across single-, bi-, and multi-temporal sequences (0.015-10m GSD), complemented by elevation and social vector layers. We comprehensively evaluate 13 frontier LLMs on our benchmark, revealing three persistent challenges: 1) disaster-domain grounding exposes unique failure modes (damage-semantic grounding, sensor-modality mismatch, and disaster-pipeline composition); 2) agents are doubly bottlenecked by tool selection and argument grounding, where gold tool-order hints improve accuracy by only 1.08-4.40%, and alternative scaffolds yield at most a 3.24% gain; 3) compositional fragility scales with trajectory length, the agent-to-gold gap widening from 7% to 56% on long pipelines. DORA establishes a rigorous testbed for operationally reliable disaster-response agents.

Rollout generation bottlenecks large-scale RL training for LLM agents. Tree rollout has emerged as an important agentic-RL strategy: by branching from shared intermediate states across reasoning turns, tool calls, and environment observations, it avoids regenerating entire trajectories from scratch and exposes long prefixes for context reuse. However, existing RL and LLM-serving infrastructure remains largely tree-unaware: it treats sibling branches as independent generation requests, making the rollout tree invisible to routing, preemption, and KV-cache management. Consequently, shared prefixes are scattered across servers, discarded with request-private suffixes, or recomputed after cross-server spill. We present Canopy, a tree-aware rollout scheduling system that makes the active rollout tree a first-class serving abstraction without changing sampling, rewards, advantage estimation, or optimizer updates. Canopy consists of three mechanisms: tree-aware routing and retention, which keep siblings near reusable prefixes and protect active prefixes; prefix-preserving partial preemption, which separates shared-prefix KV from private-suffix KV under pressure; and spill-aware transfer-or-recompute admission, which transfers shared-prefix KV to a spilled sibling only when its estimated transfer cost is lower than destination-side recomputation. On long-horizon multi-turn agent workloads, Canopy achieves up to 2.11× rollout-generation speedup; across four representative tree-rollout algorithms, mechanism studies show higher prefix locality and a 51.1% average, 94.1% maximum reduction in full-release shared-prefix invalidations. These results show that exposing rollout-tree structure to the rollout-serving stack can reduce wasted prefill and KV-cache churn, enabling more efficient RL training for LLM agents.

AI benchmarks increasingly leverage item-level statistical models, particularly item response theory (IRT), to estimate model capabilities, rank systems, select informative examples, and diagnose benchmark quality. However, AI benchmark data often departs from the data regime of human testing, for which standard IRT estimation tools were originally developed: benchmarks typically involve fewer evaluated models, far more items, and capability distributions that may be skewed, clustered, or multimodal. We examine how these regime mismatches challenge the reliability of IRT modeling for AI evaluation. Using item parameters and capability distributions derived from six widely used LLM benchmarks, we simulate response matrices under three common IRT models and compare four estimation tools used in recent benchmark studies: marginal maximum likelihood, Markov chain Monte Carlo, variational inference, and a neural pseudo-Siamese estimator. Across 18,000 simulation conditions, we systematically evaluate computational feasibility, scalability, and the reliability of IRT inferences about model rankings, predicted performance, and item characteristics. Results show that classical estimators can become infeasible in large benchmark settings, whereas scalable estimators can produce unreliable item-level and ranking inferences with small or nonnormally distributed model sets. This study identifies when latent trait models reliably support or risk distorting AI benchmarking claims, and what sample sizes and diagnostics are needed for trustworthy use.

Knowledge editing has emerged as a promising approach for efficiently updating embedded knowledge in large language models (LLMs). It first computes an ideal target hidden state that steers the LLM toward the new output, and then edits the model parameters so that the original hidden state is mapped to this target. While most existing methods focus on the second stage of parameter editing in preserving other knowledge, we find that the upstream construction of the target hidden state also plays an important role in allocating the model's limited capacity space. We first establish a theoretical connection between interference in the parameter space and interference between target hidden states, showing that the two are positively correlated. Building on this insight, we propose a simple yet effective $\ell_1$ regularization method that prunes away unnecessary ``tentacles'' of each knowledge vector, retaining only the essential components. With fewer extraneous tentacles, each edit affects less other knowledge, enabling better coexistence. Extensive experiments across multiple LLMs and benchmark datasets verify the effectiveness of our method.

Large Language Models (LLMs) are increasingly deployed in agentic contexts, where the model relies on a system prompt to use tools and complete tasks. Implicit in these agentic settings are operational requirements: using the right tools, keeping prompts at reasonable length, achieving parsimonious solution paths, and complying with safety and formatting policies. For many practitioners, assembling domain-specific supervised data to post-train LLMs to satisfy such requirements is infeasible. In this paper, we introduce CAPO (Constraint-Aware Prompt Optimization), an \emph{explicit threshold-constrained} prompt-optimization algorithm based on primal--dual updates. CAPO employs a primal--dual framework to optimize system prompts under explicit constraints via Lagrangian relaxation. Our results across three agentic benchmarks show that CAPO more reliably reaches empirically feasible operating points while improving agentic performance. We also demonstrate that the algorithm generalizes beyond agentic use cases, achieving strong performance on assistant-style evaluations that require satisfying output-format and safety/privacy constraints. Finally, we show that CAPO's rewrite policy can be amortized into DCAPO, a feedback-aware trainable rewriter that matches CAPO accuracy while improving the constrained trade-off.

Multimodal recommendation leverages item content, such as text, images, and audio signals, to complement sparse user-item interactions. Existing methods mainly combine content and collaborative signals at the representation level through graph propagation, feature fusion, or auxiliary alignment. However, continuous item embeddings make semantic sharing implicit and provide limited support for transferring behavioral evidence through reusable semantic factors. Discrete semantic codes offer explicit shared variables, but existing code-based recommenders usually treat item-code assignments as fixed outputs of pretrained quantizers, clustering algorithms, or frozen encoders, making the assignment mechanism insensitive to user behavior. We propose \textbf{CASQRec}, a \textbf{C}ollaborative-\textbf{A}daptive \textbf{S}emantic \textbf{Q}uantization framework for multimodal recommendation. CASQRec learns item-code assignments as content-inferable routing decisions supervised by collaborative behavior during training. It constructs a semantic-code graph, where fixed content-derived RQ codes are graph nodes and item-code edges are predicted from item content by a learnable prior router. To make routing behavior-aware while keeping item-code routing independent of item-side interaction history at inference time, CASQRec introduces a collaborative posterior teacher only during training and distills its assignment distributions into the content prior. The resulting prior-induced user-item-code graph allows collaborative evidence to be shared through discrete semantic codes while keeping item routing content-based. Experiments on three public datasets show that CASQRec consistently outperforms representative multimodal recommendation baselines. The code resources are available in the supplementary materials.

Causal discovery aims to uncover the causal relationships among variables beyond correlations from observational data. On time series, a major challenge is nonstationarity, which is typically modeled as regime-based shifts or drifting causal strength. In contrast, we consider non-stationarity arising from varying cause-to-effect delay, and introduce a causal model that is identifiable with few assumptions. To discover causal relationships with varying causal delay in practice, we formalize the problem in terms of the algorithmic model of causation, and propose STRETCH to discover causal graphs. We show on synthetic data that our method correctly recovers the causal relationships, and show the relevance of this causal model on a case study on the impact of El Niño climatic phenomenon on Indian monsoon.

Vision-Language-Action (VLA) models for autonomous driving produce natural-language reasoning alongside predicted trajectories, but whether this reasoning reflects the causal structure of the scene remains untested. We introduce CausalDriveBench, an evaluation framework grounded in Pearl's Causal Hierarchy (PCH) that tests causal reasoning in driving-specific VLAs through structured visual question answering (QA) and alternative-trajectory prediction. To this end, we construct causal scene graphs over nuScenes that distinguish causally active, dormant, and distractor entities, separating perceptual salience from causal relevance. The benchmark spans all four rungs of PCH (association, intervention, and counterfactual along with causal discovery) for QA generation. For the higher rungs, we additionally provide reference trajectories under specified scene modifications, enabling action-level verification that complements reasoning-level evaluation. In total, the benchmark contains 7,285 verified causal QA pairs and 1,000 counterfactual trajectories derived from nuScenes. We evaluate 10 driving-specific VLAs and 3 general-purpose VLMs, and report three findings. First, the best model reaches only 70.6\% QA accuracy, and 4 of 13 models score below random chance. Second, comparing each driving VLA to the general-purpose VLM that shares its language backbone, the cost of driving fine-tuning ranges from 2 to 34 percentage points on causal QA, with post-training design explaining the spread. Third, causal QA and trajectory accuracy are statistically uncorrelated across models: under counterfactual prompts, predicted trajectories either over-react or collapse onto the observed-scene baseline. Taken together, these results show that neither fluent rationales nor accurate observed-scene trajectories constitute evidence of causal understanding.

We study causal inference under outcome interference for sequential, observational settings. We consider settings where the binary outcomes over $N$ units are Markovian across $T$ time steps; at each time step, the outcomes of $N$ units have pairwise dependencies captured through an Ising model; and, each outcome is impacted through a latent external field capturing effects of latent confounders. Similar to panel data literature, these latent confounders are modeled to have a low-rank factor structure. Our data is a single sample from this high-dimensional distribution. To estimate causal quantities of interest, we provide a computationally efficient method based on Maximum Pseudolikelihood Estimation (MPLE) for learning the model parameters. Under reasonable assumptions, we establish non-asymptotic consistency for parameter estimation. Therefore, sampling from the learnt model enables faithful estimation of causal quantities of interest. We demonstrate the efficacy of the method through synthetic experiments as well as a real-world case-study investigating causal effects of vaccine rates on COVID-19 death rates within US counties nationwide.

Causal discovery, the problem of inferring the direction of causality, is generally ill-posed. We use the language of structural causal models (SCM) to show that assuming that the causal relations are acyclic and invariant across multiple environments (e.g., the way minimum wage affects employment rate is stable across different geographical regions), \textit{only} two auxiliary environments are sufficient to infer the causal graph for arbitrary nonlinear mechanisms. Moreover, we demonstrate that this implies identifiability of the SCM functional mechanisms: as a corollary, we show that \textit{two} auxiliary environments are sufficient to guarantee correct counterfactual inference. We empirically support our theoretical results on synthetic data.

Single-cell transcriptomics enables profiling of cellular states at unprecedented resolution, but its high dimensionality, sparsity, and technical batch effects pose significant challenges for representation learning. Existing single-cell foundation models typically encode each cell independently or only model cells from the same batch for denoising, thereby underutilizing the rich relational information across batches and cell types to model gene expression patterns. We argue that single-cell models can benefit from more informative cell-context modeling. By comparing consistency and variation across cells, models can capture fine-grained gene-gene dependencies associated with cell states, which are essential for learning high-quality representations. Inspired by the use of multiple sequence alignment (MSA) context in protein modeling, we propose \textbf{CellMSA}, a single-cell representation learning framework that introduces an MSA-inspired inductive bias into transcriptomic modeling. For each target cell, CellMSA retrieves relevant cells from different batches and biologically related cell types as context, and summarizes cross-cell patterns into a context-dependent gene-pair representation. This representation is then injected into a pair-aware target-cell encoder for fine-grained representation learning. We pretrain CellMSA on a large-scale human single-cell corpus of approximately 109 million cells. Experiments show that our framework consistently outperforms existing methods across multiple benchmarks.

Chain-of-thought (CoT) oversight is often motivated by a simple premise: if a model exposes its reasoning, then its behavior should be easier to monitor. This position paper argues that the premise conflates several properties that can come apart in practice. To make this separation explicit, we distinguish four properties: reasoning visibility (whether a trace is disclosed), mechanistic faithfulness (whether reasoning step content is supported by internal activations), behavioral monitorability (whether a specified monitor can recover a target property on held-out cases), and mechanistic localizability (whether a compact causal mechanism can be found). Claims about one require evidence specific to that property. We support this position with two mirror-image case studies on open-weight reasoning models. In security vulnerability reasoning, probes over internal activations confirm that key reasoning steps are faithfully represented, yet a reference-assisted LLM jury cannot reliably judge whether the resulting analysis is correct on held-out cases. In authority-hint factual reasoning, the pattern reverses: an external monitor detects the target behavior near-perfectly, but repeated mechanistic analyses fail to isolate a compact causal circuit. These results show that internal support does not guarantee monitorability, and monitorability does not guarantee localizability. We argue that CoT oversight papers should adopt property-specific reporting: every claim should state the reasoning access tier, target property, monitor class, monitor competence, held-out protocol, baseline, and evidence type.

Imagine Mr. Bean stepping into Tom and Jerry--can a video model generate interactions that stay faithful to each character's identity, behavior, and style across worlds? We present MiMix, a unified framework for multi-character text-to-video generation that goes beyond visual appearance to capture character-specific personas generalizing to unseen scenes and partners. To evaluate this, we introduce the Cross-Character Generalization Benchmark, a social stress test placing each character in novel scenes with previously uncoexistent partners. MiMix combines Cross-Character Embedding (CCE), which disentangles identity and behavior via text-aligned annotations, with Cross-Character Augmentation (CCA), which synthesizes cross-style training data while preserving native appearance. Experiments show consistent gains in identity fidelity, interaction quality, and style consistency over prior personalized and general video models.Additional results and videos are available on our project page: https://mi-mi-x.github.io.

As sequence lengths continue to grow, distributed attention based on sequence parallel (e.g., Ring-Attention) has become a mainstream solution for large language models (LLMs). Nevertheless, inherent communication dependencies limit computation–communication overlap, reducing GPU utilization. Moreover, in cross-hardware (NUMA or node) architectures, due to imbalanced communication bandwidth, frequent communication synchronization will limit overall performance. Based on these insights, we propose CHARM+, Cross-Hardware Attention with Re-Merge Multistream Mechanism. We reorganize sequential communications into several levels and achieve intra-level parallelism and computation overlap. Furthermore, we decouple cross-hardware communication from intra-hardware communication, eliminating synchronization overhead caused by bandwidth imbalance. Forward and backward algorithms were evaluated on multiple hardware architectures and consistently outperformed state-of-the-art methods. Compared to Ring-Attention, our forward and backward algorithms achieve average speedups of $1.3\times$ and $1.7\times$ on a dual-NUMA PCIe architecture, and $2.8\times $ and $3.9\times$ on a dual-node NVLink architecture, respectively.

Routing is foundational in VLSI physical design, directly shaping timing, power, and area. Recent pre-trained Seq2Seq models recast routing as autoregressive generation over DFS-serialized trees, but inherit a text backbone blind to the physical space their tokens describe: DFS decouples sequence position from physical coordinate, breaking spatial reasoning at branching points, in load selection, and along the autoregressive trajectory. We name these failures the \emph{Three Spatial Blindnesses} and propose ChiP-STAR, which augments a T5Gemma backbone with three lightweight modules: a dual-frame rotary encoding coupling sequence order with 3D proximity, a Fourier-factorized attention prior with memory linear in sequence length, and a cumulative coordinate-noise schedule closing the train--inference gap, together adding under $0.5\%$ parameters. We rebuild the AiEDA corpus into the largest geometry-annotated routing dataset to date ($6.4$M nets, $1.8$B tokens, $556$\,GB). ChiP-STAR-Large surpasses iPCL-Large by $5\%$ on Leaf IoU and Connectivity, and commercial ECO signoff yields $3\%$ shorter wirelength and $11\%$ fewer vias. Our code is publicly available at \url{https://anonymous.4open.science/r/iPCL-R-0778/}.

Large Language Models (LLMs) struggle with temporal alignment, often producing outdated time-sensitive facts as real-world knowledge evolves. Yet the field's ability to measure this misalignment is itself fragile. Existing benchmarks are static snapshots, rely on rigid templates, are small, lack provenance, conflate willingness-to-answer with recall, and use phrasings that newer models have likely memorised. We argue that progress on temporal alignment is bottlenecked less by data scarcity than by evaluation design, and we contribute on both fronts. We present ChronosAlign, a natural language, programmatically updatable dataset and evaluation framework for temporal alignment in LLMs; generated from a Wikidata and Wikipedia corpus of ~500,000 question-answer pairs covering ~27,000 unique questions across sport, politics, and culture from 2000 to 2025. Each question carries a Wikipedia URL and a documented update procedure, so the same evaluative claim can be re-tested against future models without re-authoring the benchmark. We couple ChronosAlign with a framework focussed on three prompting modalities, explicit (year-anchored), relative ("current" anchored), and invariant (untimed control), whose disagreement isolates which component of temporal alignment a model fails. We benchmark sixteen LLMs surfacing three findings: (i) a sharp post-2022 decay in explicit recall even for models with later training cutoffs, (ii) systematic misalignment between explicit and relative recall, with some models better aligned to explicit year prompts and others to relative anchors, and (iii) a confounding effect of refusal behaviour. Our generation and update code can be found at: https://github.com/sg-sy/ChronosAlign. Our dataset can be found at: https://huggingface.co/datasets/sg-sy/ChronosAlign.

Existing knowledge distillation methods supervise only what a student outputs, leaving how it computes those outputs unconstrained, so students may match teacher behavior through entirely different internal mechanisms. We propose Meta Circuit Distillation (MCD), which reframes distillation as the explicit transfer of reasoning circuits, the structured computational pathways uncovered by mechanistic interpretability. MCD represents each model's computation as a transcoder-based attribution graph and aligns teacher and student circuits via an optimal-transport pathway-matching loss that plugs into any standard distillation pipeline. We prove that minimizing this loss bounds the discrepancy in MLP-level computation between teacher and student. Across two model families, six distillation objectives, and three instruction-following benchmarks, MCD delivers consistent improvements that transfer to mathematical reasoning, with the largest gains under aggressive compression.

Large language models often improve reasoning by sampling multiple outputs and aggregating their final answers, but precise and efficient control of error levels remains a challenging task. In particular, deciding when to stop sampling remains difficult when the stopping rule is data-dependent and the set of possible answers is not known in advance. We study anytime-valid certification of a prespecified target answer as the unique mode of the model’s response distribution, a guarantee distinct from answer correctness. We propose the Certification by Intersection-union Testing with E-processes (CITE) algorithm, which provably controls false certification at any prescribed level under arbitrary data-driven stopping, without requiring prior knowledge of the answer category set. We also prove an category-set-size-free stopping-time rate, establish matching minimax lower bounds up to constants in the main regime, and extend the construction to confidence-weighted voting. Simulations and LLM self-consistency experiments show empirical error control and improved certification in diffuse-tail settings.

Incremental learning aims to continuously learn from dynamic data streams without suffering from catastrophic forgetting. However, existing methods may fail in challenging scenarios where class and domain spaces expand simultaneously, struggling to learn continually and stably from fragmented data streams. To address this challenge, we propose EMCift (Extensible Multi-Center Modeling with Dual-Path Drift Harmonization), a novel model that treats heterogeneous data across domains and semantic spaces as co-existing multi-view observations, enabling the construction of global knowledge from local observations. Our approach organizes extensible multi-center prototypes as a tree hierarchy, allowing the global class representation across different domains to grow organically from fragmented local data. To sustain this dynamic topology, we introduce a dual-path drift harmonization strategy that balances structural stability with adaptive plasticity. Specifically, one path captures essential updates to enable prototype evolution, while the other path leverages a conditional adversarial mechanism to robustly initialize domains, effectively preventing catastrophic forgetting caused by class confusion. Extensive experiments on three datasets demonstrate the state-of-the-art performance of our proposed model in the class-domain incremental learning scenario. Code will be released upon acceptance.

We show that AI agents are capable of discovering novel algorithms for adversarial attacks against LLMs, advancing the state of the art on white-box jailbreaking and prompt injection evaluations. We deploy frontier agents, such as Claude Code and Codex, in an autoresearch loop with access to a library of 30+ prior methods and an evaluation script with a fixed compute budget. We show this pipeline to be effective in jailbreaking OpenAI's GPT-OSS-Safeguard-20B and in prompt injections against Meta-SecAlign-70B, an adversarially robust model. For GPT-OSS-Safeguard, the best agent-discovered method achieves up to 80\% attack success rate on CBRN queries, compared to <50\% for existing methods. For SecAlign, it achieves 100\% ASR, while the best prior automated methods only achieve 82\%. Notably, in our setting, attack methods are developed on unrelated surrogate models for a pure random-target token-forcing task, yet generalize directly to prompt injection on the adversarially trained model. Finally, we trace the lineage of methods developed during autoresearch, characterizing the agents' strategies and failure modes. Adversarial ML has long held that defenses must be evaluated against attacks tailored to them; autoresearch automates this principle, and we argue it should be the minimum bar for defense evaluation going forward.

With the rapid advancement of text-to-image diffusion models, increasingly realistic AI-generated content has raised serious concerns about misuse and copyright infringement. Digital watermarking offers a promising solution by enabling user-level traceability, yet existing methods remain difficult to deploy at scale, hindered by costly model retraining or detection, poor effectiveness–fidelity trade-offs, and limited robustness to image transformations. To address these challenges, we propose CLaW (Codec-guided Latent Watermarking), a robust and efficient watermarking framework for traceable diffusion image generation with frozen diffusion backbones and inversion-free detection. Specifically, CLaW first uses a pretrained watermark codec to map each watermark message to a latent residual, enabling low-cost sampling-time injection. Then, the residual is injected within a late denoising window, where image semantics are better preserved and the watermark signal is less perturbed by subsequent denoising updates, yielding a better balance between watermark effectiveness and image fidelity. Furthermore, we introduce a decoder-guided adaptive injection mechanism that uses decoding confidence as feedback to dynamically calibrate watermark strength, reinforcing weak watermark signals and improving robustness under image transformations. Extensive experiments show that CLaW can preserve visual fidelity and improve average F1 under image transformations, achieving a 6.34% gain over the state of the art.

Large language models are increasingly deployed as autonomous agents for multi-step workflows in real-world software environments. However, existing agent benchmarks are limited by trajectory-opaque grading, underspecified safety and robustness evaluation, and narrow coverage of modalities and interaction paradigms. We introduce Claw-Eval, an end-to-end evaluation suite addressing these gaps with 300 human-verified tasks spanning 9 categories across three groups: general service orchestration, multimodal perception and interaction, and multi-turn professional dialogue. To enable trajectory-aware grading, each run is recorded through three independent evidence channels: execution traces, audit logs, and environment snapshots, yielding 2,159 fine-grained rubric items. The scoring protocol evaluates Completion, Safety, and Robustness, with Average Score, Pass@k, and Pass^k across three trials to distinguish genuine capability from lucky outcomes. Experiments on 14 frontier models show that: (1) Trajectory-opaque evaluation is systematically unreliable, missing 44% of safety violations and 13% of robustness failures detected by our framework. (2) Capability does not imply consistency, with Pass@3 remaining stable under error injection while Pass^3 dropping by up to 24 percentage points. (3) Agent capability is strongly multi-dimensional, with model rankings varying across task groups and metrics, indicating that our heterogeneous evaluation coverage is essential. Claw-Eval highlights directions for developing agents that are not only capable but reliably deployable.

Style transfer aims to render target content in the style of a reference image, but existing methods often suffer from content leakage, where objects, layouts, or semantics from the style reference appear in the generated output. Although prior data-driven and training-free methods can reduce leakage, they often face a leakage--degradation dilemma: stronger content suppression may weaken style fidelity, while richer style preservation may reintroduce unwanted reference content. We identify this dilemma across the full style-transfer pipeline, including feature separation, feature-space grounding, and diffusion generation. To address these issues, we propose CLeaR, a training-free framework for content-leakage-resistant style transfer. CLeaR first uses Orthogonal Subspace Projection to define content-reduced style targets in each vision foundation model (VFM) feature space. It then performs Ensemble Inversion, which optimizes a shared pixel-space style anchor satisfying style constraints across multiple VFMs. Finally, Energy-Guided Calibration maintains style alignment during diffusion sampling by steering the denoising trajectory toward the ensemble-defined style manifold. We further provide a theoretical analysis showing that the style-anchor estimation error decreases with the number of VFMs. Experiments on StyleBench demonstrate that CLeaR improves style alignment, reduces content leakage, and achieves better LLM-as-Judge evaluation compared with existing methods.

This paper studies the problem of semi-supervised edge classification, which aims to identify edge relations using both labeled and unlabeled data. This problem is highly challenging due to the inherent asymmetry of edge relations and strong prediction biases from imbalanced homophilous and heterophilous edges. Towards this end, this paper proposes a novel approach named Complementary Tripartite Play with Bayesian CaLibration (CLEAR) for semi-supervised edge classification. The core of our CLEAR is to incorporate asymmetric semantic branches and a meta-teacher into a tripartite-play framework, enabling reliable guidance and optimization under label scarcity. In particular, our CLEAR introduces a topological branch and a contextual branch which extracts graph semantics, generating pseudo-labels in complementary views. To improve the reliability of pseudo-labels, we filter low-quality pseudo-labels using epistemic uncertainty, and then calibrate posterior distributions using the priors induced by neighborhood information. More importantly, we utilize a meta-consoler to guide the optimization of two branches by outputting the objective coefficients, ensuring a reliable pseudo-labeling process in a tripartite play framework. Extensive experiments on benchmark datasets validate the superiority of the proposed CLEAR in comparison to various baselines. Our code is available at https://anonymous.4open.science/r/CLEAR-D1D7/.

Recovering dynamical systems from noisy observations is a recurring challenge across scientific domains, including neuroscience and physics. Latent stochastic differential equations (SDEs) address this by modeling the system as an unobserved state that evolves according to a learnable SDE and generates the observations. Variational inference (VI) provides a tractable objective for fitting latent SDEs. Traditional VI algorithms evaluate this objective by numerical simulation over a time discretization, trading fidelity for computational cost. A recent class of algorithms, simulation-free VI, sidesteps this tradeoff by parameterizing the posterior through its instantaneous marginals rather than its drift. In this work, we show that the efficiency of simulation-free VI algorithms comes at a price: their parameterizations restrict the approximate posterior to a subset of the SDEs available to simulation-based methods, degrading both posterior inference and parameter learning. We propose Helmholtz-SDE, a simulation-free VI algorithm that closes this gap by optimizing over path laws compatible with a prescribed collection of marginals. Helmholtz-SDE recovers dynamics more faithfully than prior simulation-free methods, with the largest gains under high posterior uncertainty. It further matches the performance of simulation-based VI at a fraction of the runtime.

Traditional rendering pipelines rely on complex assets, accurate materials and lighting, and substantial computational resources to produce realistic imagery, yet they still face challenges in scalability and realism for populated dynamic scenes. We present C2R (Coarse-to-Real), a generative rendering framework that synthesizes real-style urban crowd videos from coarse 3D simulations. Our approach uses coarse 3D renderings to explicitly control scene layout, camera motion, and human trajectories, while a learned neural renderer generates realistic appearance, lighting, and fine-scale dynamics guided by text prompts. To overcome the lack of paired training data between coarse simulations and real videos, we adopt a two-stage synthetic-real domain-hedging strategy that first learns a strong generative prior from large-scale real footage, then introduces controllability by using a small amount of paired synthetic coarse-fine data to anchor shared implicit spatio-temporal features across domains. The resulting system supports coarse-to-fine control, generalizes across diverse CG and game inputs, and produces temporally consistent, controllable, and realistic urban scene videos from minimal 3D input.

GUI agents interact with environments by perceiving interfaces and executing actions. As a virtual sandbox, the GUI World model empowers agents with human-like foresight by enabling action-conditioned prediction. However, existing text/pixel-based approaches struggle to achieve both high visual fidelity and fine-grained structural controllability. To this end, we propose \textbf{Code2World}, a vision-language coder that simulates the next visual state via \textbf{renderable code generation}. Specifically, to address the data scarcity problem, we construct \textbf{AndroidCode} by translating GUI trajectories into high-fidelity HTML and refining synthesized code through a visual-feedback revision mechanism, yielding a corpus of \textbf{over 80K} high-quality screen-action pairs. To adapt existing VLMs into code prediction, we first perform SFT as a cold start for format layout following, then further apply \textbf{Render-Aware Reinforcement Learning} which uses rendered outcome as the reward signal by enforcing visual semantic fidelity and action consistency. Extensive experiments demonstrate that Code2World-8B achieves the top-performing next UI prediction, rivaling the competitive GPT-5 and Gemini-3-Pro-Image. Notably, \textit{Code2World significantly enhances downstream navigation success rates in a flexible manner}, boosting Gemini-2.5-Flash by {+9.5\%} on AndroidWorld navigation.

Cross-modal knowledge distillation transfers knowledge from a teacher modality to a student modality. Existing feature-level alignment methods typically assume that teacher and student features reside in structurally alignable representation spaces. However, this assumption does not hold when cross-modal features are structurally heterogeneous and lack clear unit-level correspondence, such as 2D spatial visual grids and 1D temporal audio sequences, thereby limiting the applicability of feature-level alignment. To address this challenge, we propose a cross-modal distillation framework that enables effective knowledge transfer across structurally heterogeneous feature spaces via a vector-quantized codebook. Specifically, teacher features are abstracted into a set of vector-form codes regardless of their original feature structure, and the selected codes serve as concept-level anchors for student learning. Code selection is guided by both task relevance and student compatibility, allowing the student to receive transferable teacher knowledge without requiring direct unit-level feature alignment. Experimental results across diverse cross-modal distillation scenarios demonstrate the effectiveness of the proposed framework on classification and semantic segmentation tasks.

Large language models have achieved remarkable capabilities across diverse domains, yet their safety alignment remains vulnerable to jailbreak attacks. In this work, we identify a previously underexplored failure mode—safety generalization lag—where alignment trained predominantly on natural language fails to transfer to the code domain. We show that this lag induces a code-completion blind spot, allowing malicious intent embedded within syntactically valid code to evade safety mechanisms. To exploit this vulnerability, we propose CodeMimicry, a fully automated black-box jailbreak framework that generates structured, object-oriented code prompts to induce harmful outputs via code completion. Experiments on 8 state-of-the-art commercial LLMs demonstrate that CodeMimicry achieves a 96.25\% attack success rate with 1.51 queries on average, significantly outperforming both template-based and optimization-based baselines. Beyond empirical performance, we provide a mechanistic analysis of code-based jailbreaks through latent space representations, including projection onto refusal-related directions and activation steering. This analysis offers an explanation of how CodeMimicry bypasses safety mechanisms in code-related domains. Our findings reveal a weakness in current safety alignment and highlight the need for robust alignments in structured domains such as code.

As AI agents ship more code with minimal human oversight, code review becomes the critical quality-control checkpoint — but one without a reliable, objective verifier: review quality is nuanced and ultimately a judgment call by the PR author. Existing benchmarks are structurally misaligned with this: annotators are third-party reviewers, like the bots they evaluate, not the PR authors who actually decide what to fix. They also rely on small, static issue sets that require full re-evaluation as tools update, limiting how frequently progress can be tracked. We introduce CodeReviewBench, built on a fundamentally different signal: revealed developer preferences — if a developer acts on a bot's suggestion, we treat it as useful; if they ignore it, we do not. This signal is mined daily from the public GitHub event stream, yielding 500K+ judged PRs across 18 tools, and refreshes continuously as tools evolve. We validate the developer preference against an independent offline benchmark: the two agree on precision rankings and independently detect the same product updates within days of vendor releases. Together, these form the benchmark.

Clinical Decision Making (CDM) requires integrating heterogeneous information across sequential diagnostic stages while interacting with patients whose behaviors are often non-stationary and unreliable. However, existing medical agents are typically trained against cooperative patient simulators and tackle each clinical task in isolation, making them fragile to dynamic patient behaviors and unable to capture inter-task dependencies essential for globally coherent decision-making. To address these issues, we propose CoE-Agent, a co-evolving patient–doctor agentic framework that learns globally coherent CDM strategies via interactive policy graph optimization. First, Dynamic Patient–Doctor Co-Evolutionary Learning pairs a Dynamic Patient Agent, which maintains an evolving cognition state to simulate non-stationary behaviors, with an Interactive Policy Graph-Based Doctor Agent that generates DAG-structured policy graphs unifying interactions, tool invocations, and reasoning steps. The doctor agent is then optimized via Interactive Policy Graph Optimization (IPGO), which jointly enforces structural validity and task-aligned correctness through reinforcement learning. Second, Cross-Task Policy Graph Consolidation and Unified Model Learning prunes low-value nodes, aggregates high-value nodes across tasks while filtering redundancy, and distills a unified policy planner with task-specific decision models for globally coherent decision-making. Experiments on MedChain and ClinicalBench show that CoE-Agent surpasses state-of-the-art baselines by 11.13\% and 23.63\% in average score, while achieving up to 19.5$\times$ faster inference. Source code is to be released.

As LLM agents advance beyond short-horizon tasks, evaluating their ability to perform long-horizon strategic decision-making in realistic multi-agent settings remains a key challenge. Existing agent benchmarks typically focus on isolated task completion or interactions with rule-based or fixed-policy counterparts, limiting their ability to capture strategic adaptation and emergent behaviors arising from sustained multi-agent interactions. We introduce \textbf{CoffeeBench}, a benchmark for evaluating LLM agents in a dynamic multi-agent economic environment consisting of farmers, roasters, and retailers forming a multi-stage supply chain. The environment consists of autonomous firms interacting over a multi-month horizon with evolving supply, demand, and pricing, creating sustained competition and supply-chain dynamics. In each evaluation, the evaluated model controls one firm and must make sequential decisions on procurement, production, pricing, and negotiation to maximize cumulative net income. CoffeeBench enables the study of emergent behaviors arising from repeated strategic interactions, including negotiation dynamics, pricing discipline, and coordination failures. We evaluate several frontier and pareto-frontier LLMs on CoffeeBench to analyze their economic performance and strategic behavior. Most evaluated models achieve positive net income, with GPT-5.5 achieving the highest mean net income. In contrast, Claude~Haiku~4.5 frequently yields negative net income, exhibiting an idle-drift failure mode in which agents collapse into inactivity despite coherent plans. We further find that stronger models proactively negotiate with counterparties through frequent messaging, whereas weaker models trade reactively with limited communication. CoffeeBench provides a controllable testbed for studying strategic interaction, coordination, and long-horizon behavior in multi-agent economies.

Reasoning capabilities are critical for advancing Large Language Models, yet current approaches either require massive computational budgets or struggle to effectively distill reasoning to smaller models. Standard distillation methods rely on outcome-based rewards, failing to distinguish between sound reasoning and lucky guesses. We propose Collaborative Reasoning Distillation (CRD), a framework that enhances reasoning in compact models through three innovations: (1) interactive cross-feedback where teachers iteratively critique each other's reasoning, (2) fine-grained step-wise quality assessment capturing logical validity independent of final answers, and (3) coherence-aware step stitching that synthesizes complementary strengths. Students are trained via Reasoning Quality Optimization (RQO) with budget constraints. Our model, CRD-4B, achieves 97.3\% on MATH-500 and 70.3\% on AIME'25, surpassing baselines while using only 50K training examples, up to 12 times smaller than the datasets of comparable models.

The Hessian of the energy with respect to the nuclear positions is indispensable in atomistic modelling. However, constructing this matrix requires $O(N)$ Hessian vector products, traditionally limiting high-accuracy Hessians to small systems. Machine learning interatomic potentials (MLIPs) have accelerated atomistic modelling by providing highly accurate energies and forces at $O(N)$ cost, yet the resulting $O(N^2)$ cost of Hessians remains a practical bottleneck for large systems. Based on the insight that we can derive the sparsity pattern for an MLIP's Hessians in closed form, we show in this paper how to use techniques from sparse automatic differentiation to reduce the cost of a local MLIP's Hessians to a system-size-independent number of Hessian-vector products, yielding overall $O(N)$ total cost without any approximations. We benchmark our approach on a variety of systems ranging from alkane chains to water clusters to A$\beta$40 conformers. Depending on the MLIP configuration, we achieve the linear scaling regime already on relatively small systems, resulting in large runtime reductions between $2\times-15\times$ for these systems. This opens up the possibility of scaling high-accuracy MLIP Hessians to very large systems, such as proteins that were previously inaccessible.

On the simplest non-Euclidean manifold, the 1D flat torus, we exhibit a distribution where a recently proposed extension of Riemannian flow matching is globally optimized by a velocity field that pushes samples away from the data distribution. We trace this failure mode to a property we call compatibility; a likelihood-based flow matching objective is compatible if its population minimizer recovers the marginal velocity field induced by the data and coupling. We give a general condition for compatibility, apply it to existing methods, and bound the worst-case error in terms of a residual component vanishing under compatibility. Next, we note that existing compatible objectives put density on velocities that are infeasible on compact manifolds. We propose an objective built from an exponential family over endpoints, with the conditional velocity as sufficient statistic, which is both compatible and supported only on feasible velocities. This manifests as a re-weighting of the flow matching loss which, on constant-curvature manifolds, treats errors in flow speed and direction differently. Empirically, our method recovers the correct velocity field on the test case, outperforms prior methods on high-dimensional product tori, and obtains substantial NLL improvements over prior work on two of four standard spherical geospatial benchmarks at moderate runtime overhead.

Prefill-only KV compression freezes a token subset at the end of prefill and decodes from it without further eviction. The retention decision is therefore irreversible, yet existing methods estimate the corrective signals it relies on, per-head reliability and prompt-level compression sensitivity, online from a single noisy prompt. We argue this is the wrong statistical unit: these signals exhibit far higher cross-prompt regularity than within-prompt signal-to-noise. We introduce CompilerKV, a KV-retention policy whose corrective tables are compiled offline from a calibration corpus, reducing online correction after the standard observation-window scan to $O(1)$ lookups plus a budget clamp. We find that compiled retention tables behave as portable architectural priors: rankings transfer across disjoint corpora on four backbones, with mean Spearman $\bar{\rho}=0.90$, and direct model-to-model table transfer costs only $0.4$-$0.8$ LongBench points on average. At a 512-token budget, CompilerKV attains compressed-SOTA on all four backbones, improving over the strongest prefill-only baseline by $+1.67$ points on average, with task-bootstrap 95% CI $[+1.08,+2.37]$. Pressure regimes amplify the gap: under a fixed $512/32\mathrm{k}$ cache ratio, CompilerKV remains the strongest compressed method through 128k RULER, with approximately 73 versus FullKV 79 and SnapKV 38; on 32k NIAH it reaches $0.89$ versus SnapKV $0.42$; and at 32k input, retaining only $1.56\%$ of the prefill KV, batch-16 serving remains feasible where FullKV is OOM.

Post-training compression of Transformer models commonly relies on truncated singular value decomposition (SVD) strategies leading to low-rank approximations of the original model weights. However, the underlying single shared subspace modeling can degrade accuracy even at moderate compression. Sparse dictionary learning provides a more flexible union-of-subspaces representation, but existing approaches often suffer from costly iterative dictionary and coefficient updates. We propose COMPOT (Calibration-Optimized Matrix Procrustes Orthogonalization for Transformers), a training-free compression framework that uses a small calibration dataset to estimate a sparse weight factorization. COMPOT employs orthogonal dictionaries that enable closed-form Procrustes updates for the dictionary and analytical single-step sparse coding for the coefficients, reducing significantly the overall optimization cost. To handle heterogeneous layer sensitivity under a global compression budget, COMPOT further introduces a one-shot dynamic allocation strategy that adaptively redistributes layer-wise compression rates. Extensive experiments across diverse architectures and tasks show that COMPOT consistently delivers a superior quality–compression trade-off over strong low-rank and sparse baselines, while remaining fully compatible with post-training quantization for extreme compression. Code will be released upon acceptance.

Competition among LLM providers hinges not only on model accuracy, but also on the latency of serving user requests. However, user demand leads to congestion which increases latency, creating a feedback loop between provider decisions and user choices. In this work, we study a stylized game between two providers who strategically balance allocating fixed compute between training better models and reducing latency. A population of users then chooses between the providers. Our equilibrium analysis reveals that a lower-budget provider can still attract users by taking advantage of the congestion faced by the dominant provider. Relative to a competitive baseline where providers have equal compute resources, compute asymmetry leads the higher-resource provider to divert a greater fraction of their compute from improving accuracy to reducing latency. Moreover, this compute asymmetry strictly harms users. Altogether, our results illustrate how congestion distorts competition between model-providers, leading to subtle implications for market competitiveness, model accuracy, and user utility.

Neural surrogates for Partial Differential Equations (PDEs) are usually trained on synthetic data from numerical simulators. As models grow and scaling laws demand larger datasets, the cost of data generation often outweighs the training cost and becomes the bottleneck in the scaling quest. Unlike data in other domains, however, each PDE sample has a tunable per-sample cost that depends on its numerical fidelity, with the resulting cost-accuracy trade-off governed by known convergence rates. We exploit this property by formalizing neural surrogate training as a function of both data quantity and fidelity under a fixed data generation compute budget, treating fidelity as a fine-grained discrete spectrum rather than a binary distinction. Building on this, we show that multi-fidelity training enables Pareto-optimal error scaling against data generation compute. By using simulations at lower fidelities as an additional training signal, our approach consistently outperforms training on the highest fidelity at matched data generation budgets. We validate this across six PDE problems spanning structured grids and unstructured meshes, two distinct error sources (discretization and iterative-solver error), and four surrogate architectures.

Multimodal Large Language Models (MLLMs) trained on sequential tasks suffer from catastrophic forgetting. We study this problem in LoRA-based multimodal continual instruction tuning. By analyzing the diagonal Fisher information of LoRA parameters, we find that forgetting is amplified not merely by shared high-sensitivity parameters across tasks, but by the concentration of updates on those shared parameters, \ie, some tasks distribute their change uniformly across parameters, while others pack most of it into a narrow subset, causing disproportionate interference on the parameters that prior tasks depend on. We further observe that this concentration manifests in the dominant directions of recent gradients. Based on this insight, we propose \textbf{D}om\textbf{i}nant-direction \textbf{G}radient \textbf{Pro}jection (\textbf{DiGPro}), which maintains a short buffer of recent gradient directions, extracts their dominant subspace with adaptive rank selection, and attenuates the parallel component for each optimizer step. Crucially, DiGPro requires no prior-task data or stored subspaces, operating on the current gradient history. Experiments on cross-modal and vision-centric continual learning benchmarks show DiGPro consistently reduces forgetting and attains competitive final results. The code is available in the supplementary materials.

Few-shot whole-slide image (WSI) classification remains challenging due to gigapixel-scale images, scarce slide-level labels, and strong morphological heterogeneity. Recent vision-language MIL methods improve data efficiency by transferring knowledge from pathology foundation models, but they often treat patches as weakly related instances and represent each class with a single semantic or visual anchor. We propose a multi-scale vision-language framework that combines prompted pathology encoding, pathology concept learning, Semantic Wasserstein Routing (SWR), and barycentric prototype memory. Using frozen pathology vision-language backbones with lightweight prompt adaptation, our method aligns image patches with concept embeddings through unbalanced optimal transport and builds a semantic patch graph for concept-guided aggregation. In parallel, the prototype memory maintains multiple class-specific Wasserstein barycenters to capture diverse morphological modes. A distance-based fusion strategy integrates slide-level text alignment, concept-guided patch evidence, and prototype-memory matching for robust prediction. Experiments on TCGA WSI classification benchmarks under few-shot settings show consistent improvements over strong MIL and vision-language MIL baselines, demonstrating the benefit of semantic patch routing and multi-prototype reasoning for weakly supervised pathology learning.

Three-dimensional Gaussian Splatting (3DGS) has recently become a promising representation for dense SLAM, due to its explicit scene parameterization, efficient differentiable rendering, and high-quality visual reconstruction. Extending 3DGS to semantic SLAM enables the construction of maps that are not only geometrically accurate but also semantically interpretable. However, existing semantic Gaussian SLAM methods usually optimize semantic, depth, and normal renderings through separate supervision objectives. Such a design underexplores the fact that these outputs are generated from the same Gaussian field and should therefore provide a concordant explanation of the underlying scene structure. This limitation is especially problematic around object boundaries and geometric discontinuities, where inaccurate Gaussian growth can cause cross-boundary expansion, boundary blurring, and semantic leakage. To address this problem, we propose Concord-SLAM, a boundary-native semantic Gaussian SLAM framework that establishes concordance among semantic, depth, and normal renderings while promoting boundary-aware Gaussian growth. Our framework contains two key components. First, Cross-Render Concordance (CRC) jointly examines semantic, depth, and normal views rendered from the same Gaussian map, and converts their structural discrepancies in boundary responses, regional continuity, and local geometry into optimization signals. In this way, disagreement among different renderings is used as an intrinsic cue for improving the semantic-geometric coherence of the Gaussian field. Second, Boundary-Native Splatting (BNS) actively inserts new Gaussians near semantic boundaries, depth discontinuities, and salient geometric variations, encouraging primitives to align with physical and semantic boundaries during map construction rather than correcting boundary mixing only through later optimization. Experiments on Replica and ScanNet demonstrate that Concord-SLAM achieves superior performance in camera tracking, scene reconstruction, and semantic mapping.

Ranking models define probability distributions over rankings and are fundamental in applications such as preference learning and decision-making. Without structural assumptions, these models are intractable due to their factorial complexity. Although independence assumptions are key to reducing complexity and enabling interpretability, standard notions of conditional independence do not directly apply to rankings because they are subject to mutual exclusivity constraints. We introduce \emph{relative rank models}, a probabilistic framework based on structured coarsenings of the ranking space that relax these constraints while preserving relative order information. This representation enables a well-defined notion of conditional independence for rankings and allows rankings to be modeled using Bayesian networks. As a result, relative rank models support tractable learning, probabilistic inference, and interpretable representations. We further show that two prominent ranking models based on independence assumptions tailored to rankings arise as special cases, providing a unifying framework for independence in ranking models.

Conformal Prediction (CP) provides a model-agnostic framework for constructing finite-sample prediction sets with coverage guarantees and has become a tool for reliable decision-making in high-risk settings. Recent advances have extended Split Conformal Prediction to increasingly complex learning problems, including multilabel, high-dimensional, and functional outputs. In this work, we study CP for Distribution-to-Distribution Regression, a setting of growing interest across many application domains. We show that a naive application of Split CP yields prediction sets that, while marginally valid, are non-interpretable, e.g., lacking a well-defined volume or a tractable sampling procedure, and of limited practical value. To overcome this, we propose a new conformal framework that operates in an orthogonal basis space, producing interpretable prediction sets while preserving coverage guarantees. We further develop an adaptive variant with asymptotic conditional coverage under mild assumptions. Empirical results on synthetic and real-world datasets validate the effectiveness of the proposed approach.

Federated Learning (FL) enables collaborative model training across clients while preserving data privacy. However, FL typically relies on voluntary client participation with uniform rewards, which can lead to high-quality clients dropping out due to inadequate incentives and low-quality clients taking a free-ride, both of which degrade overall model performance. However, existing incentive mechanisms fail to capture the true contribution value of a client’s data to the global model, and often skew client selection toward cost compression rather than quality enhancement, thereby exacerbating confounding bias. To address this, we propose CausalAFL, a causal-based framework that reduces information asymmetry and corrects for confounding factors. Specifically, CausalAFL builds a complete causal chain connecting client information, bid, and marginal contribution. By jointly optimizing an inference network and a generative network within a variational inference framework, it precisely adjusts for causal bias. Maximizing the evidence lower bound objective allows CausalAFL to separate true contribution from noise and highlight the value of high-quality clients. A social welfare objective that quantifies both natural direct and indirect effects guides the server to prioritize these clients, resulting in improved global model accuracy and social welfare. Experiments on multiple datasets demonstrate that CausalAFL can guide clients to adjust their bids, leading to winners with higher contributions, confirming its effectiveness.

Bayesian model-based spatial clustering methods are widely used for their flexibility in estimating latent clusters with an unknown number of clusters while accounting for spatial proximity. Many existing methods are designed for clustering finite spatial units, limiting their ability to make predictions, or may impose restrictive geometric constraints on the shapes of subregions. Furthermore, the posterior clustering consistency theory of spatial clustering models remains largely unexplored in the literature. In this study, we propose a Spatial Domain Random Partition Model (Spat-RPM) and demonstrate its application for spatially clustered regression, which extends spanning tree-based Bayesian spatial clustering by partitioning the spatial domain into disjoint blocks and using spanning tree cuts to induce contiguous domain partitions. Under an infill-domain asymptotic framework, we introduce a new distance metric to study the posterior concentration of domain partitions. We show that Spat-RPM achieves a consistent estimation of domain partitions, including the number of clusters (which may go to infinity), and derive posterior concentration rates for partition, parameter, and prediction. We also establish conditions on the hyperparameters to achieve consistency, offering important practical guidance for hyperparameter selection. Finally, we examine the asymptotic properties of our model through simulation studies and apply it to Atlantic Ocean data.

Federated Learning (FL) enables collaborative model training without centralizing private data, but its performance often degrades under data heterogeneity because local updates become biased toward client-specific data characteristics. In this paper, we propose Constant Term Shrinkage for Federated Learning (FedCTS)}, a simple server-side aggregation method designed to mitigate such update drift. FedCTS decomposes each client update into a constant term and a residual term. We show that the constant term is strongly affected by data heterogeneity. FedCTS therefore evaluates the reliability of the constant term and adaptively shrinks the unreliable constant term toward zero while leaving the residual term unchanged. The shrinkage strength is obtained in closed form from clients' update statistics, without requiring an additional tuning hyperparameter. Experiments across multiple datasets and FL algorithms demonstrate that FedCTS improves robustness under heterogeneous data with negligible additional computation, no extra communication overhead, and strong compatibility with existing FL techniques.

Generative models, including diffusion models, are increasingly used as foundation models and adapted through sequential fine-tuning, making continual learning an essential problem setting. However, continual learning in such generative models remains poorly understood: after a task change, what aspects of the learned distribution are most easily lost, and what replay samples should be prioritized? We address these questions through the modern Hopfield energy. Recent links between modern Hopfield networks (MHNs) and diffusion models allow analyses in MHNs to be transferred to diffusion models. We introduce intrinsic forgetting as an increase in Hopfield energy after the task change. In tractable settings in an MHN, we prove that high-energy, outlier-like samples undergo a larger energy increase than cluster-like samples, implying that samples located in sharp, isolated basins are more forgettable. We further analyze memory replay and show that replay is particularly effective for high-energy samples, enabling an energy-based selection of replay samples. We validate these predictions in experiments on MHNs and two diffusion models under continual-learning settings: Stable Diffusion and a pixel-space DDPM. In these diffusion models, Hopfield energy tracks reconstruction-based forgetting, and replay experiments reveal energy-dependent mitigation of forgetting that is consistent with the MHN analysis.

Large audio language models (LALMs) have shown impressive capabilities on diverse audio understanding tasks, ranging from speech transcription to music analysis. However, because LALMs are typically trained to produce text-aligned responses, their hidden states are progressively shaped for text generation rather than for preserving acoustic information. As a result, the diverse acoustic content that audio carries, such as phonetic detail, prosody, sound events, affect, and pitch, is lost along the way and difficult to leverage in the response. We introduce Continuous Audio Thinking (CoAT), a framework that equips audio language models with a continuous latent workspace for organizing acoustic information prior to response generation, grounded by distillation from audio experts. Within the thinking space, the model can utilize the rich acoustic information provided by expert distillation when generating its response. Furthermore, the proposed continuous thinking block can be processed in a single prefill, so CoAT does not require additional autoregressive decoding cost over the baseline. Across three LALMs, Qwen2-Audio, Qwen2.5-Omni-7B, and Audio Flamingo~3, performance gains on a broad benchmark suite spanning audio reasoning, audio understanding, music classification, speech emotion, and speech transcription demonstrate the effectiveness of CoAT. Further analysis confirms that the auxiliary supervision propagates from the thinking positions to the model's textual responses.

Deep Gaussian processes (DGPs) are appealing Bayesian regression models, but their standard discrete-depth parameterization often becomes harder to optimize as depth grows. We argue that depth is better treated as a latent continuous evolution than as a long discrete stack. We therefore propose a continuous-depth formulation in which the latent representation follows a stochastic differential equation with a depth-indexed drift family equipped with a GP-based prior. This reframes the core design choice as the prior placed on that family, rather than adopting a purely neural continuous-depth parameterization as in neural ODE or neural SDE models. We study two instantiations of this idea: CDGP, a direct GP prior over the state-depth drift, and FlowDGP, a flow-evolved prior that captures richer depth dependence at lower training cost. To motivate the redesign, we show that the single-sample DGP objective can become increasingly ill-conditioned under repeated layer composition, while residual DGP is a useful first repair that still remains a finite discrete construction. We also give pathwise sensitivity analysis for the continuous-depth formulation. Experiments on synthetic, benchmark, and large-scale regression tasks show that residual DGP improves over plain DGP, while CDGP and FlowDGP provide a stronger overall story across the regimes we study.

Autoregressive language models have achieved remarkable success in text modeling, but recent work increasingly challenges the fixed left-to-right generation order. However, existing non-autoregressive alternatives such as discrete diffusion, still struggle to jointly deliver efficiency, scalable representation learning, and global semantic modeling. We propose Cola DLM, a hierarchical continuous latent diffusion language model that learns a stable Text VAE, models a global semantic prior in latent space with a block-causal DiT, and decodes text conditionally. Theoretically, under a unified Markov-path formulation, the diffusion operates latent prior transport rather than token-level observation recovery, thereby decoupling global semantic organization from local token realization. Empirically, Cola DLM exhibits strong scaling behavior for text generation across four research questions, eight benchmarks, strictly matched $\sim$2B-parameter AR and LLaDA baselines, and scaling curves up to $\sim$2000 EFLOPs. Meanwhile, we also explore encoding text in the same continuous latent space as images, providing a feasible path toward unified discrete-continuous generative modeling. Our code will be released.

A core quality of general intelligence is the ability to open-endedly expand and evolve its set of mastered skills autonomously. While recent Foundation Model (FM) driven approaches have shown promising results towards this goal, they typically rely on significant human-in-the-loop engineering, limiting their transferability to novel environments. To address this, we introduce Continuous Open-ended Discovery and Evolution of Skills as Hierarchical Reward Programs (CODE-SHARP), a framework that leverages FMs to open-endedly grow and evolve an archive of Python programs encoding skills to train a generalist agent policy entirely from scratch via reinforcement learning, directly from source code. These programs, termed Skills as Hierarchical Reward Programs (SHARPs), each encode a local success condition and a set of prerequisites delegated to previously discovered SHARPs. At runtime, SHARPs dynamically route the agent through their prerequisite chain based on the current state, rewarding each completion along the way, requiring the agent to learn only the marginal behaviour each new SHARP introduces, enabling efficient learning of long-horizon skills without any pre-defined rewards. On Craftax-Classic and XLand, agents trained fully autonomously by CODE-SHARP outperform previous works by 6x and 2.6x in median performance and are the only agents capable of crafting iron tools and mining diamonds. Scaled to Craftax-Extended, CODE-SHARP trains a generalist agent on over 90 discovered SHARPs, enabling the agent to solve challenging long-horizon tasks zero-shot, matching agents trained on ground-truth rewards.

Existing conditional diffusion models typically inject condition information into the reverse-time denoising process, either as model inputs or through guidance. Although effective for conditional fidelity, this design burdens the denoiser when compact continuous signals, such as user preference ratios or mixture degrees, must support both image reconstruction and condition-wise separation, often degrading sample quality. In this work, we reinterpret this limitation not as a problem of condition representation, but as a problem of condition placement. We propose the Continuous Personalized Diffusion Model (CPDM), which shifts the locus of conditioning from reverse-time signal injection to forward geometry formation by coupling bounded spinor-component coordinates with normalized image-space drift directions. The resulting condition-dependent forward drift forms condition-specific terminal geometry, from which continuous changes along the spinor-component coordinate are reflected as continuous visual transitions rather than endpoint interpolation. Experiments show that, even with a single scalar condition, CPDM achieves competitive performance compared with standard and higher-capacity reverse-conditioning baselines and supports continuous generation under compact scalar control. These results suggest forward geometry design as a practical alternative to conventional reverse-time conditioning for compact continuous personalization.

In the classical *identification in the limit* model of Gold [Inf. Control, 1967], a stream of positive examples is presented round by round, and the learner must eventually recover the target hypothesis. Recently, Kleinberg and Mullainathan [NeurIPS 2024] introduced *generation in the limit*, where the learner instead must eventually output novel elements of the target's support. Both lines of work focus on positive-only or fully labeled data. Yet many natural supervision signals are inherently relational rather than singleton: comparative experiments, A/B tests, side-by-side judgments, and similarity–dissimilarity annotations produce observations that encode relationships between examples rather than labels of individual ones. This motivates us to initiate the learning-theoretic study of *contrastive identification and generation in the limit*, where the learner observes a *contrastive presentation* of data: a stream of unordered pairs $\\{x,y\\}$ satisfying $h(x)\ne h(y)$ for an unknown target binary hypothesis $h$, but which element is positive is hidden from the learner. We first present three results in the noiseless setting: an exact characterization of contrastive identifiable classes (a one-line geometric refinement of Angluin's tell-tale condition [Angluin, Inf. Control 1980]), a combinatorial dimension called *contrastive closure dimension* extending the closure dimension of Raman et al. [COLT 2025] and exactly characterizing uniform contrastive generation with tight sample complexity, and a strict hierarchy in which contrastive generation and text identification are mutually incomparable. We then prove a sharp *reversal* under finite adversarial corruption: there exist classes identifiable from contrastive pairs under any finite corruption budget by a single budget-independent algorithm, yet not identifiable from positive examples under even one corrupted observation. The unifying technical object is the *common crossing graph*, which encodes pairwise ambiguity, family-level generation obstructions, and corruption defects in a single coverage-and-incidence language.

Large reasoning models separate inference into an explicit thinking phase followed by a final answer phase, yet how the answer phase uses the reasoning trace remains unclear. We present a systematic study of this trace-to-answer transition across multiple LRMs and find that (i) final answers diverge from correct reasoning traces at rates between 6\% and 48\%, even when traces terminate cleanly at full budget, (ii) naively copying the trace answer underperforms standard decoding in some settings, revealing that the answer phase has independent corrective capacity, and (iii) the same phenomenon extends beyond structured tasks to settings where reasoning traces and final answers can diverge under social-pressure framing (sycophancy). These findings motivate Contrastive Thinking Decoding (CTD), a training-free, single-model decoding method that contrasts answer-phase logits under the primary reasoning trace against those from a deliberately degraded noisy trace. CTD selectively amplifies trace-consistent tokens while preserving the answer phase's corrective capacity, concentrating CTD-internal joint mass on the both-correct cell (T$\checkmark$A$\checkmark$) without inducing trace-correct-to-answer-wrong drift relative to the base distribution. Across math reasoning, science, code, social-pressure resistance (sycophancy), and a broad knowledge benchmark (MMLU), \ctd reduces trace-answer disagreement and improves the accuracy--compute trade-off without parameter updates or auxiliary models at decoding time.

Generative agents have proven to be powerful assistants in a wide variety of contexts. Given this success, users are now deploying agents with minimal restrictions in open ended, multi-agent environments. Current methods for measuring the dynamics of open-ended multi-agent systems are limited to qualitative inspection. In this paper, we extend the process-theoretic notion of adaptive control charts to multi-agent systems to enable automated monitoring. Using simulation, we demonstrate that adaptive control charts are necessary for monitoring multi-agent systems that can learn from their environment. We further demonstrate, both empirically and theoretically, that adaptive control charts are susceptible to adversarial agents that defect sufficiently slowly. These results illustrate a fundamental tradeoff in multi-agent system control: either agents in a system cannot learn \textit{or} the system is susceptible to adversaries.

Method validation and study design in causal inference rely on synthetic data with known counterfactuals. Existing simulators trade off distributional realism, the ability to capture mixed-type and multimodal tabular data, against causal controllability, including explicit control over overlap, unmeasured confounding, and treatment effect heterogeneity. We introduce CausalMix, a variational generative framework that closes this gap by coupling a mixture of Gaussian latent priors with data-type-specific decoders for continuous, binary, and categorical variables. The model incorporates explicit causal controls: an overlap regularizer shaping propensity-score distributions, alongside direct parameterizations of confounding strength and effect heterogeneity. This unified objective preserves fidelity to the observed data while enabling factorial manipulation of causal mechanisms, allowing overlap, confounding strength, and treatment effect heterogeneity to be varied independently at design time. Across evaluation settings, CausalMix achieves strong distributional fidelity on mixed-type tables while providing stable, fine-grained causal control. We demonstrate practical utility in a comparative safety study of metastatic castration-resistant prostate cancer treatments, using CausalMix to compare estimators under calibrated data-generating processes, tune hyperparameters, and conduct simulation-based power analyses under targeted treatment effect heterogeneity scenarios.

The rapid growth of video-based social media has increased users’ exposure to harmful content, creating a need for reliable automated video safety detection. Although recent Vision-Language Models (VLMs) show strong video understanding capabilities, existing harmful video detection systems face two key limitations: they typically reduce safety detection to binary classification, overlooking the inherently multi-label nature of unsafe videos, and they rely on static training objectives that do not support controllable precision-recall trade-offs, though the desired operating point may vary across moderation pipelines and unsafe categories. To address these gaps, we propose Adaptive Tversky Policy Optimization (ATPO), a reinforcement learning framework for Multi-label Video Safety Detection (Multi-VSD). ATPO introduces the Adaptive Tversky Reward (ATR), which dynamically adjusts false-positive and false-negative penalties during training to enable controllable precision–recall trade-offs. Experiments on SafeWatch-Bench and XD-Violence show that ATPO substantially improves multi-label performance, increasing the Jaccard Index from 40.66 to 75.44 on SafeWatch-Bench-Real. Moreover, ATR enables reliable steering of the precision–recall operating point, supporting deployment scenarios with heterogeneous policy requirements.

A convergence analysis is developed for the regularized Newton method for training neural networks (NNs) in the overparameterized limit. We prove that as the number of hidden units tends to infinity, the NN training dynamics converge in probability to the solution of a deterministic limit equation involving a "Newton neural tangent kernel" (NNTK). Explicit rates characterizing this convergence are provided and, in the infinite-width limit, we prove that the NN converges exponentially fast to the target data (i.e., a global minimizer with zero loss). Crucially, we show that this convergence is uniform across the frequency spectrum, addressing the spectral bias inherent in gradient descent. The eigenvalues of the neural tangent kernel for gradient descent accumulate at zero, leading to slow convergence for target data with high-frequency components. In contrast, the NNTK has uniformly lower bounded eigenvalues if the regularization parameter is selected appropriately, allowing Newton's method to converge more quickly for data with high-frequency components. Mathematical challenges that need to be addressed in our analysis include the implicit parameter update of the Newton method with a potentially indefinite Hessian matrix and the fact that the dimension of this linear system of equations tends to infinity as the NN width grows. This substantially complicates deriving the training dynamics in the overparameterized limit as well as proving the convergence of the finite-width dynamics thereto. The analysis identifies a scaling formula for selecting the regularization parameter, which we show can vanish at a suitable rate as the number of hidden units becomes larger. In addition, we prove that, for sufficiently large numbers of hidden units, the regularized Hessian remains positive definite during training and the Newton updates for individual NN parameters converge to zero, demonstrating that the model behaves as a linearization around the initialization.

Discovering cooperative behavior in multi-agent reinforcement learning (MARL) is challenging due to the combinatorial complexity of joint state-action spaces, which hinders the emergence of coordinated behaviors from trial-and-error alone. Intrinsic rewards are often used to aid discovery, but naively combining them with team objectives can distort the learning signal, compromising task performance. In this paper, we propose EFXPLORER, a constrained exploration framework that maximizes exploration objectives subject to a constraint that preserves established task performance. We solve it via an epigraph reformulation that introduces adaptive exploration budgets. This approach separates intrinsic rewards from task objectives and regulates exploration through task feasibility. To further encourage diverse and temporally extended exploration, we incorporate a successor distance-based intrinsic reward that captures long-horizon dependencies. Empirically, our method outperforms strong baselines and induces novel cooperative strategies across SMAX, VMAS, and MPE benchmark suites.

Parameter-Efficient Fine-Tuning (PEFT) commonly adapts pretrained weights through low-rank updates, and recent methods further exploit the singular value decomposition (SVD) of the base weight for initialization or subspace selection. However, these methods do not explicitly preserve the coupled geometry between the pretrained left and right singular bases. Motivated by recent minimum-perturbation theory, which shows that stable finetuning admits a coherent SVD rotation in which a single orthogonal $Q$ acts on both the left singular basis $U_0$ and the right singular basis $V_0$, we prove a per-slice analogue: each row slice of $W_0$ can be adapted by a shared orthogonal rotation $Q_i$ on its left basis $U_i$ and right basis $V_i$ together with a diagonal spectrum shift. We instantiate this form as **CORA** (*Coherent Orthogonal Rotation Adaptation*), which applies per-slice orthogonal rotations and a per-layer diagonal scale to the rank-$r$ SVD truncation of $W_0$. CORA uses $\tfrac{1}{2}m(r{-}1)$ trainable parameters per linear layer, about $4{\times}$ fewer than LoRA at the same rank. CORA outperforms LoRA, DoRA, PiSSA, and MiLoRA on commonsense reasoning and code generation while using $\sim 8{\times}$ fewer parameters.

Large Vision-Language Models (LVLMs) are prone to hallucinations, producing responses that are inconsistent with visual inputs. While inference-time activation steering provides a lightweight solution without retraining, existing methods are either limited by coarse global steering vectors or rely on unconstrained full-dimensional vector prediction, which can produce unreliable vector for unseen hallucination patterns. In this paper, we propose \textbf{Correction Space Steering} (CSS), an robust activation steering framework that predicts correction coordinates in a learned low-dimensional correction space. We first conduct an empirical analysis of hallucination-related activation differences and find that they are predictive of hallucination types, form type-aware geometric structures, and exhibit low effective ranks in most layers. We further provide a theoretical analysis showing that these activation differences can be approximated by a low-dimensional subspace. Based on this insight, CSS learns a low-dimensional correction space from activation differences and trains an online predictor to infer input-specific correction coordinates from inference-time hidden states. The resulting correction vectors are then reconstructed within the correction space and applied to steering activations for hallucination mitigation. Experiments on standard hallucination benchmarks show that CSS effectively reduces hallucinations without retraining LVLMs and outperforms existing inference-time mitigation methods.

Intracranial electrocorticography (ECoG) offers high–signal-to-noise access to cortical activity for brain–computer interfaces, yet limited per-patient data has led most prior work to rely on small, subject-specific decoders that neglect information shared across patients. We investigate whether large pretrained scalp-EEG foundation models (EEG FMs) can be adapted to ECoG, enabling cross-patient learning and competitive decoding performance while calibrating to a held-out patient in $10$--$30$ minutes on a single GPU. We introduce CORTEG, a cross-modality transfer framework that combines a pretrained EEG FM backbone, an electrode-aware KNNSoftFourier spatial adapter, a dual-stream tokenizer for low-frequency and high-gamma activity, and a leave-one-subject-out fine-tuning strategy. We evaluate CORTEG on two challenging regression tasks: public finger trajectory regression (n=9) and private audio envelope regression (n=16). CORTEG matches or exceeds the strongest task-specific baselines on both tasks: it reaches the highest mean correlation among compared methods on the public finger benchmark (gain not statistically significant on $n{=}9$ subjects), with larger and statistically significant gains on the audio task and in low-data per-patient calibration. Feature analyses align with neurophysiology, and latent manifolds capture low-dimensional finger-movement structure. CORTEG provides systematic evidence that scalp-EEG pretraining can be repurposed for ECoG decoding, enabling data-efficient intracranial BCIs that can adapt to new patients.

Decoding cognitive functions and brain diseases from functional magnetic resonance imaging (fMRI) is a foundational problem in neuroscience advanced by recent deep learning across task and resting-state (rs) settings. Existing region-level decoders capture spatial structure through region-of-interest (ROI)-aware graph or Transformer architectures, but apply a uniform temporal operator across all regions. Consequently, they do not explicitly model how cortical regions integrate information over heterogeneous timescales. We propose CoRTeX, a region-resolved recurrent architecture for fMRI in which each cortical region carries its own learnable parameters and evolves on its own learned timescale. At its core, per-region differentiable temporal integration windows allow each ROI to learn how far back in time to integrate information through a soft mask over a private buffer. Cross-region information is restricted to a single identity-aware attention channel, additionally biased by a co-activation-driven memory whose learning and forgetting rates are themselves region-specific. Experiments on stimulus classification using task-fMRI (NSD) and on two disease diagnosis tasks using rs-fMRI (ABIDE, COBRE) demonstrate that CoRTeX achieves competitive performance with strong baselines while enabling region-level analyses that are not well-defined for models with exchangeable units: the learned per-region timescales exhibit interpretable spatial organization, and class-specific regional attribution aligns with category-selective cortex. Code is available at: https://anonymous.4open.science/r/CoRTeX-F.

Chain-of-thought (CoT) prompting and related strategies that elicit intermediate reasoning traces have fundamentally changed how language models are used, evaluated, and developed. CoT now touches many areas of contemporary research beyond prompting, from advanced test-time inference and model tuning to interpretability, yet much of this work treats CoT traces in task-specific ways, without a shared formal account of what they are or how they compose and support inference. We argue that probabilistic programs — ordinary programs extended with stochastic choices and probabilistic conditioning — provide a natural framework for modeling CoT reasoning. To demonstrate this, we derive a small discrete probabilistic programming language for CoT in which reasoning steps are stochastic choices, branches encode dependencies, trace likelihoods score executions, and answers are return values. Even with a minimal reachability-based semantics, this language enables a first-principles analysis of trace likelihood, exposing why likelihood alone can be a misleading model of reasoning. Motivated by this limitation, we add differentiable operators for composing scores and feedback over CoT traces. The resulting framework links several uses of CoT — scoring, evidence aggregation, feedback propagation, model tuning, and step-level sensitivity analysis — within a single probabilistic program. Through case studies, we show how this approach yields new analysis techniques and test-time inference strategies, offering a promising direction for programming-language-based approaches to language-model reasoning.

Online conformal prediction can fail when predictions shape actions and actions determine which outcomes enter calibration. Standard adaptive methods may retain marginal coverage while systematically miscovering the counterfactual outcomes of rarely selected actions. This paper formalizes the failure through counterfactual coverage and introduces Propensity-Weighted Online Conformal Prediction, an inverse-propensity-weighted recursion that debiases calibration. A doubly robust variant further reduces nuisance bias to the product of outcome-model and propensity errors. Under positivity, the resulting coverage rate matches an information-theoretic lower bound up to logarithmic factors. Experiments on synthetic decision tasks, open bandit data, and financial rebalancing show that PW-OCP and DR-OCP improve counterfactual coverage and downstream regret without sacrificing prediction-set sharpness.

We investigate the problem of learning neural operators for coupled multi-physics PDEs, where multiple physical processes co-evolve and influence one another via bidirectional coupling. For example, in subsurface CO$_2$ storage, fluid flow, heat transfer, and geo-mechanical deformation evolve in a tightly coupled manner. Prior work, including neural operators and generative PDE solvers, mostly focuses on single-physics settings or relies on iterative inference, limiting their effectiveness in coupled PDEs. This paper introduces coupled-flow, a one-step generative neural operator that revisits coupled operator learning as solving a distributional transport problem. It parameterizes process-specific average velocity fields conditioned on the full coupled state, enabling cross-process interactions to emerge naturally from the transport dynamics. We further establish a Wasserstein-2 generalization bound that ties the prediction error of coupled-flow to its training losses through the regularity of the induced flow. Experiments on a wide variety of multi-physics benchmarks demonstrate that coupled-flow achieves significant performance improvements over coupled neural-operator and generative PDE baselines.

Guidance methods are essential tools for steering the generation process in diffusion models, improving the generalizability of stochastic robotic controllers and enabling high-quality controlled image synthesis. Despite strong parallels between flow-matching and diffusion, integrating guidance into flow-matching models remains quite challenging. Recent work has proposed general guidance frameworks for flow-matching, however, they are direct adaptations of diffusion guidance methods and can prove unstable. In this work, we conduct a theoretical analysis of guidance in flow-matching and show that a naive translation of diffusion-based guidance to flow-matching breaks the coupling distribution used during training, resulting in weaker control over sampling accuracy. Building on this insight, we propose Coupled Guidance, a flow-matching-specific guidance mechanism derived from a geometric perspective that leverages the coupling structure inherent to flow-matching training. To validate the effectiveness of our approach, we conduct extensive experimentation on a custom robotic collision avoidance task using Maniskill3, four robotic dexterous manipulation tasks from the Adroit benchmark, and three image generation tasks, namely inpainting, super-resolution, and deblurring, using the ImageNet-128 and CelebA-HQ-256 datasets. Our findings show that our framework improves sample quality and guidance stability across all tasks, without imposing any additional overhead.

Reinforcement learning (RL) for large language models (LLMs) is increasingly bottlenecked by rollout cost, making low-precision rollout appealing for acceleration. However, applying low-precision formats such as FP4 to RL remains unstable, even with quantization-aware training (QAT) enhancements. We identify that the root cause is that FP4’s aggressive quantization amplifies small system-level numerical differences (i.e., training-inference stack mismatch) that are harmless at BF16 into large token-level log-probability errors, distorting importance weights and destabilizing optimization. To address this, we propose Cross-Precision Alignment (CPA), a simple regularizer that stabilizes low-precision RL by maintaining a BF16 master policy, executing rollouts in real FP4, and aligning the fake-quantized forward pass with a BF16 reference forward pass on sampled tokens by applying a low-variance KL penalty at each training step. In our evaluation, the resulting BF16 checkpoint matches a full BF16 RL baseline, and the overall pipeline delivers 8–19\% end-to-end training speedup.

Cyclic peptides are attractive next-generation therapeutics. In drug discovery, diverse ring-closing chemistries are applied to tune synthetic accessibility and conformational constraints. Current machine learning methods represent cyclization through modified positional encodings, covalent-bond prompts in all-atom models, residue-level distance guidance, or direct terminal closure by molecular mechanics (MM). These approximations provide limited control over the raw geometry, especially for non-natural linkers, and often yield unreliable structures. Here, we introduce CPSea2, a composable framework that abstracts cyclization as terminal-amide geometry constraints. We construct CPSea2-Cap (CPcap), a conformer library covering eight cyclization types, and CPSea2-Base (CPbase), a large-scale dataset of nearly 900 million mined linear peptide--protein complexes. Matching terminal-amides between CPcap and CPbase directly derives cyclic peptide--protein complexes with diverse terminal topologies. Built on CPbase, PepAny generates target-conditioned peptide binders with controllable terminal-amide geometry, expanding the accessible space beyond mined structures. Multi-target evaluations show improved raw-output plausibility over bond-prompted and distance-guided baselines. CPSea2 provides a novel framework and data foundation for flexible and diverse cyclic peptide binder design. Datasets and codes are available at https://anonymous.4open.science/r/CPSea2-6CF5/.

Reinforcement learning with verifiable rewards (RLVR) post-trains language models on multi-step reasoning by assigning a single outcome reward uniformly across all tokens in a trajectory, regardless of which steps contributed to success or failure. Improving credit assignment can address this limitation by enabling targeted refinement of faulty reasoning steps, rather than updating entire trajectories uniformly. Resets are one such simple mechanism, enabling more precise credit assignment by returning to an intermediate state and resampling its continuation, so that outcome differences can be attributed to decisions made at that point. We propose two such methods: Random-Reset Policy Optimization (RRPO), where reset states are drawn randomly from reasoning steps, and Self-Reset Policy Optimization (SRPO), where the model self-localizes the erroneous step in an incorrect trajectory and resets there. We analyze these methods within a Conservative Policy Iteration (CPI) framework. We extend it with a credit-assignment oracle that targets improvable states, defined as those whose advantage exceeds a threshold, and show that it yields provable improvements over random resets. Across models and reasoning benchmarks, SRPO consistently outperforms standard GRPO and RRPO by sampling multiple suffix continuations at a self-localized reset and learning from their rewards, using only the model itself with no external supervision.

Cross-modal scene and instance retrieval aims to retrieve a corresponding 3D environment or object given query 2D images, serving as a foundational mechanism for spatial reasoning in robotics and AR/VR. Existing methods achieve this by compressing the entire 3D map into a single global feature vector, which discards the fine-grained local geometrical structure in indoor scenes. In this work, we show that scene- and instance-level retrieval capabilities emerge from supervision on sparse, local 2D-3D correspondences, without any explicit global labels. Specifically, we propose $\textbf{CrossWeave}$, a novel framework for cross-modal retrieval that learns sparse patch-point alignment, which alone outperforms global encoding methods. Additionally, an optimal transport-based aggregator is applied over the aligned features to attain further gain by clustering structurally redundant regions, freeing descriptor capacity for semantically distinct local features. CrossWeave achieves state-of-the-art performance on both scene- and instance-level retrieval on ScanNet and 3RScan with a $\textit{single model}$, whereas existing methods require separate models for each task. Our results suggest that whole-scene encoding is not necessary for cross-modal retrieval, and that sparse local correspondence supervision is a more effective and flexible alternative.

Deep learning has accelerated automated atomic model building from cryo-EM maps, yet progress remains limited by the lack of large, high-quality datasets and standardized evaluation benchmarks. We address this gap by introducing CryoAtlas, a rigorously curated dataset and unified benchmark for density-guided atomic model building. By integrating and filtering EMDB and PDB entries (up to March 2026), CryoAtlas provides 17,809 map--model--sequence entries, including a curated 16,657-entry training split. From this resource, we construct a 618-target benchmark with a shared evaluation protocol that jointly assesses structural accuracy, model-to-map fit, and stereochemical validity. Using this framework, we benchmark state-of-the-art learning-based methods against Phenix and AlphaFold~3, comprehensively analyzing the impact of map resolution, sequence length, protein category and sequence identity on modeling accuracy. As a case study, retraining the ModelAngelo CNN component on CryoAtlas improves performance on the benchmark. CryoAtlas provides an open, reproducible foundation for future cryo-EM atomic model building research.

Message Passing Neural Networks (MPNNs) commonly suffer from over-squashing and over-smoothing issues, which limit their performance on graph-level tasks. To this end, we propose a novel Crossed Subtree Convolutional Network (CSCN) for graph classification. Specifically, CSCN first introduces an ordered subtree alignment strategy to decompose the original graph into an ordered sequence of structured subtrees. This sequence preserves rich local connectivity patterns and maintains the permutation invariance of the graph representation. Based on this, we design a crossed subtree convolution. It utilizes a sliding window to enable information interactions across different subtrees while simultaneously aggregating node features within each subtree. This design captures local information effectively while also modeling long-range dependencies. In this way, CSCN avoids deep message passing over the global graph topology and thus alleviates both over-squashing and over-smoothing. Experimental results on multiple graph classification benchmarks show that CSCN achieves competitive performance.

Sparse-Linear Attention (SLA) combines sparse and linear attention to accelerate diffusion models and has shown strong performance in video generation. However, (i) SLA relies on a heuristic split that assigns computations to the sparse or linear branch based on attention-weight magnitude, which can be suboptimal. Additionally, (ii) after formally analyzing the attention error in SLA, we identify a mismatch between SLA and a direct decomposition into sparse and linear attention. We propose CSLA, which introduces (I) a learnable router that dynamically selects whether each attention computation should use sparse or linear attention, (II) a more faithful and direct sparse-linear attention formulation that uses a learnable ratio to combine the sparse and linear attention branches, and (III) a sparse + low-bit attention design, where low-bit attention is introduced via quantization-aware fine-tuning to reduce quantization error. Experiments show that on video diffusion models, CSLA can achieve 97\% attention sparsity and deliver an 18.6$\times$ attention speedup while preserving generation quality.

Coding agents built from large language models increasingly inspect and modify stateful services. Before such an agent chooses or explains a patch, it needs a basic engineering skill: predicting how a local code or state change will affect future service behavior. CTE-Bench isolates this simulator-fidelity question from agent policy success. Each item gives the model service source, a factual interaction prefix, a post-prefix source patch or hidden-state overwrite, and a fixed suffix query schedule; an executable oracle supplies the counterfactual response trace. CTE-Bench-Core-v1 contains 255 counterfactual items over six deterministic Python services and 10,200 scored suffix decisions per model configuration. The main score is effect-step value-match (VM): exact response equality restricted to the 2,476 suffix positions where the intervention changes the oracle response. DeepSeek V4-Lite, Kimi K2.5, and Claude Sonnet 4.6 reach 60.7%, 61.5%, and 58.4% effect-step VM under teacher-forced one-step prompting, where each prediction sees earlier oracle suffix responses; hiding those responses while preserving the factual prefix reduces them to 11.4%, 28.9%, and 25.0%. As an additional probe, we run self-conditioned free-rollout on DeepSeek and Kimi; effect-step VM is 27.1% and 33.2%, and exact whole-trace match is at most 1.2%. These probes show that CTE-Bench helps identify dependence on oracle suffix feedback and compounding error in self-conditioned service simulation. We release Core-v1, the executable oracle, metadata, and evaluation scripts, with an evaluation card tying each supported claim to its memory protocol and metric.

Cross-Domain Recommendation is essential for enabling cross-sell in multi-service platforms; however, limited user overlap and strict cross-service data-sharing constraints often result in little to no ground-truth supervision for cross-domain learning. To bridge this gap, we propose an LLM-based framework for cross-domain recommendation in the non-overlapping user scenario. First, we propose an agentic pipeline that constructs cross-domain pseudo supervision, which is unavailable in the non-overlapping user scenario, by leveraging a user’s source-domain history to generate a target-domain item recommendation. Using this dataset, we further propose \underline{C}oupled \underline{U}ser-G\underline{R}ouped R\underline{E}inforcement Learning (CURE), which induces coupling between in- and cross-domain gradient updates via user-level joint normalization, thereby calibrating synthetic cross-domain updates against reliable in-domain signals. We provide a theoretical analysis showing that user-level coupling stabilizes policy updates under imperfect pseudo supervision and improves cross-domain generalization. Experiments on Amazon Reviews and MovieLens show consistent gains over baselines in cross-domain recommendation, while also improving in-domain performance. An online A/B test demonstrates a statistically significant 56\% lift in click-through rate over existing ML methods.

Reconstructing patient-specific cyclic cardiac motion at arbitrary phases is central to dynamic cardiac assessment, underpinning measurements such as ejection fraction, stroke volume, and regional myocardial deformation. Existing methods---pair-wise registration, time-conditioned decoders, and coordinate-based implicit fields---share a common implicit assumption: the cardiac cycle is treated as a sequence of discrete phases, with temporal structure stitched together through learned mappings supervised at the observed frames. Yet the cycle is, by physical fact, a closed periodic process. We propose \textbf{CycleSpectra}, a spectral motion representation that takes the cyclic structure as a starting point of the representation rather than as a property to be learned. Given a pair of sparsely sampled phase observations, the model learns the spectral structure of the underlying cyclic cardiac motion in a low-dimensional latent space, and recovers the displacement at any phase through a fixed analytic synthesis layer---time never enters the network as a learnable signal, only as a coordinate the readout takes by construction. To support evaluation, we collect a thin-slice phase-resolved 4D cardiac CT cohort and benchmark under both canonical and non-canonical input protocols, reporting whole-cycle motion-fidelity metrics alongside standard image and anatomical metrics. Under arbitrary two-frame input, CycleSpectra achieves the lowest ejection-fraction error and highest ventricular volume--time curve correlation among recent baselines, while remaining competitive on image fidelity. Beyond two-frame inference, the proposed framework supports test-time multi-frame latent fusion for higher accuracy, validated on both the in-house cohort and a cross-modality 4D cardiac MR cohort. Code will be available.

We introduce cyclic denoising—repeated forward and reverse diffusion at controlled noise amplitudes—as an extraction attack for image diffusion models. Inspired by random organization in disordered solids, where cyclic mechanical perturbations anneal the system into increasingly stable configurations, cyclic denoising exposes regions of the learned distribution that remain largely inaccessible to standard sampling. We find that this dynamics drives samples toward attractors with a broad stability spectrum, with the deepest attractors exhibiting ultrastability: they can be regenerated from near-total corruption and sustained through thousands of noising-denoising cycles. Many of these deep attractors correspond to memorized training images, including stock photographs, brand watermarks, and web-crawl artifacts. Our extraction attack requires sampler-level control, including mid-process noise injection, but no gradients and no weight inspection. Crucially, it requires no prior knowledge of training data, captions, or prompts. In contrast, prior generate-and-filter attacks commonly rely on prompted generation using known or suspected training captions, followed by large-scale sampling and post-hoc similarity or membership-inference filtering. While cyclic denoising can also be applied with prompts, our main protocol is fully unconditioned. We demonstrate the phenomenon in Stable Diffusion v1.4, a large latent diffusion model, and in a smaller pixel-space DDPM, showing consistent behavior across latent- and pixel-space diffusion models. Across noise amplitudes, we observe a dynamical transition from trivial fixed points to structured memorized images, hierarchical partial absorption in which coarse scene layout freezes while fine details remain diffusive, basin hopping between memorized states with long residence times, prompt-stabilized memorized templates, and cross-initial-condition universality of the recovered attractor set. Together, these results establish cyclic denoising as both a physics-inspired probe of generative landscapes and a practical tool for memorization auditing, with implications for privacy, copyright compliance, and model fingerprinting.

Data-centric training—selecting, mixing, and reweighting data during LLM optimization—has produced a rapidly growing body of algorithms, yet every new method ships as a bespoke codebase with its own interface, training protocol, and evaluation setup, making head-to-head comparison and reproducible evaluation nearly impossible. We present DataFlex, the first unified benchmark and evaluation platform for data-centric training of LLMs. DataFlex contributes: (i) a standardized evaluation protocol that runs selection, mixture, and reweighting methods under identical seeds, data splits, hardware, and metrics; (ii) a reference implementation of ten representative algorithms (six data selection, three data mixture, one data reweighting) re-engineered on top of LLaMA-Factory to share model, dataloader, and distributed-training infrastructure (including DeepSpeed ZeRO-3 with full-rank gradient acquisition); and (iii) an empirical benchmark whose head-to-head results yield reproducible findings that no single existing codebase could produce: dynamic selection methods consistently outperform static full-data training on MMLU across Mistral-7B and Llama-3.2-3B, and the gain is larger on smaller models (up to 13.3 points); DoReMi and ODM yield complementary wins on high- vs. low-resource domains when pretraining Qwen2.5-1.5B on SlimPajama at 6B and 30B tokens; and DataFlex's implementation is consistently faster than the original codebases of LESS, TSDS, and MoE-SFT. By unifying fragmented methods under a single evaluation substrate, DataFlex provides the community with a reproducible, extensible basis for future data-centric research.

Training and evaluation in multi-channel imaging (MCI) remains challenging due to heterogeneous channel configurations arising from varying staining protocols, sensor types, and acquisition settings. This heterogeneity limits the applicability of fixed-channel encoders commonly used in general computer vision. Recent Multi-Channel Vision Transformers (MC-ViTs) address this by enabling flexible channel inputs, typically by jointly encoding patch tokens from all channels within a unified attention space. However, unrestricted token interactions across channels can lead to feature dilution, reducing the ability to preserve channel-specific semantics that are critical in MCI data. To address this, we propose Decoupled Vision Transformer (DC-ViT), which explicitly regulates information sharing using Decoupled Self-Attention (DSA), which decomposes token updates into two complementary pathways: spatial updates that model intra-channel structure, and channel-wise updates that adaptively integrate cross-channel information. This decoupling mitigates informational collapse while allowing selective inter-channel interaction. To further exploit these enhanced channel-specific representations, we introduce Decoupled Aggregation (DAG), which allows the model to learn task-specific channel importances. Extensive experiments across three MCI benchmarks demonstrate consistent improvements over existing MC-ViT approaches.

We propose D-DOIT (Discrete Doob-Oriented Inference-time Transformation), a training-free and efficient adaptation method for discrete diffusion models with generic rewards. D-DOIT formulates adaptation as sampling from a reward-tilted target distribution and realizes this transport through Doob's h-transform of the discrete diffusion reverse kernel, using only reward values rather than reward gradients. Unlike continuous diffusion, masked discrete diffusion samples categorical token-reveal transitions rather than continuous state updates. D-DOIT derives the corresponding discrete Doob's h-transform, which guides sampling by reweighting reverse transition probabilities instead of adding a drift correction. To make this transformation practical, D-DOIT avoids expensive future rollouts. At each guided step, D-DOIT samples candidate next states, uses the model prediction head to complete each candidate into a clean sequence, evaluates each completion with the reward oracle, and resamples the next state with probabilities proportional to the rewards. An optional late-stage best-of-$K$ refinement further improves sample quality by branching trajectories only near the end of denoising, avoiding the $K$-fold cost over the full trajectory. Empirically, on regulatory DNA sequence design benchmarks, D-DOIT consistently outperforms training-free guidance baselines and achieves performance competitive with training-based methods, improving both enhancer activity and cell-type-specificity while preserving sequence naturalness.

Deadline-Constrained Cost-Aware Dynamic Workflow Scheduling (D-CADWS) aims to minimize virtual machine (VM) rental cost while maintaining high workflow deadline satisfaction in dynamic cloud environments. The problem is challenging because the deadline impact of assigning a workflow task to a VM is not directly observable at decision time. It depends on long-horizon effects, including downstream task dependencies, VM queueing, and interactions among dynamically arriving workflows. Existing methods usually fold deadline violations into a single reward or penalty term, which obscures the distinction between action safety and action cost. In this paper, we cast D-CADWS as a representation learning problem. The core idea is to learn a deadline-aware representation that can predict, before execution, whether a candidate action is risky with respect to the current task deadline, and then use it to guide cost-efficient scheduling. Based on this idea, we propose a representation-centered deep reinforcement learning (RCDRL) method. RCDRL constructs task-level deadline supervision, trains a predictive deadline model to learn deadline-aware representations, and learns a VM-cost critic on top of this representation. The final policy follows a safe-first rule that rejects risky actions before minimizing VM cost. We further propose a two-phase training strategy to keep the learned representation aligned with the evolving policy-induced data distribution. Experiments on dynamic workflow scheduling benchmarks show that RCDRL achieves substantially lower VM cost than other state-of-the-art heuristic and DRL baselines while maintaining strong workflow deadline success.

Random sampling is a fundamental tool in modern machine learning and numerical linear algebra for reducing the computational cost of large-scale matrix problems. Existing analyses, however, rely primarily on subspace embedding guarantees, which do not precisely characterize the statistical bias of nonlinear random oblique projections induced by sampling, which arises ubiquitously in subsampled least squares and fast low-rank approximation methods. Because (pseudo)inversion is nonlinear, these random oblique projections can be systematically biased even when the underlying sketch is unbiased, thereby introducing hidden bias into downstream least squares and low-rank approximation solutions. In this work, we develop a unified non-asymptotic theory for random oblique projections in high dimensions. We show that standard random sampling schemes generally induce a systematic statistical bias overlooked by classical subspace embedding-style analyses, and we propose a principled debiasing framework to correct it. We illustrate the power of the theory through two canonical applications. For subsampled least squares, we obtain sharp bias--variance characterizations, reveal previously unrecognized statistical suboptimality in widely used sampling schemes, and identify when debiasing yields provable improvements. For fast CUR decomposition, we develop a debiased approach with improved approximation accuracy. Numerical experiments further validate our theoretical findings.

The process of discovery requires active exploration---the act of collecting new and informative data. However, efficient autonomous exploration remains a major unsolved problem. The dominant paradigm addresses this challenge by using Reinforcement Learning (RL) to train agents with intrinsic motivation, maximizing a composite objective of extrinsic and intrinsic rewards. We suggest that this approach incurs unnecessary overhead: while policy optimization is necessary for precise task execution, employing such machinery solely to expand state coverage may be inefficient. In this paper, we propose a new approach that explicitly decouples exploration from policy optimization and bypasses RL entirely during the exploration phase. Our method uses a tree-search strategy inspired by the Go-With-The-Winner algorithm, paired with a measure of uncertainty to systematically drive exploration. By removing the overhead of policy optimization, our approach explores an order of magnitude more efficiently than standard intrinsic motivation baselines on hard exploration benchmarks. Further, we demonstrate that the trajectories discovered during exploration can be distilled into deployable policies using existing supervised backward learning algorithms, achieving state-of-the-art performance by a wide margin on Montezuma's Revenge, Pitfall!, and Venture without relying on domain-specific knowledge. Finally, we demonstrate the generality of our framework in high-dimensional continuous action spaces by solving the MuJoCo Adroit dexterous manipulation and AntMaze tasks in a \emph{sparse-reward} setting, directly from image observations and without expert demonstrations or offline datasets. To the best of our knowledge, this has not been achieved before.

Scaling laws have driven remarkable performance breakthroughs in Computer Vision (CV) and Natural Language Processing (NLP) by increasing model depth. Compared to increasing width, deeper networks provide higher parameter efficiency and more expressive representations. However, increasing network depth in reinforcement learning (RL) still yields limited performance gains. In this paper, through theoretical and empirical analysis, we reveal that this performance degradation is primarily attributed to implicit low-rank bias and overfitting. While these two issues also exist in shallow networks, the detrimental effects caused by component coupling are significantly amplified as the network depth increases, leading to severe performance degradation. For the first issue, we theoretically prove the existence of bootstrapped spectral coupling, which causes high-frequency spectral components to couple with the low-frequency spectral components of the value network, thereby driving representations of deep value networks into severe rank collapse. For the second issue, we reveal that deep value networks severely overfit the noise induced by TD target couplings, which means the construction of TD targets is influenced by other components, such as the boundary of value distribution or current policy. Therefore, our insight is that decoupling is the key to successfully scaling deep value networks in RL. Motivated by this insight, we propose two simple yet effective solutions respectively: policy-independent spectral decoupling and boundary-target decoupling. Moreover, decoupling policy and value learning (as in AWR) is also a necessary target decoupling for deep value network training in offline RL. By integrating these decoupling components, we propose Spectrally Decoupled Distributional Learning (SEED). Empirically, to the best of our knowledge, SEED is the first approach to successfully unlock the potential of depth in both online and offline RL settings, enabling value networks to scale effectively from 3 to 32 layers with consistent performance gains.

Many real world processes can be represented as compositions of functions along a Directed Acyclic Graph (DAG). In causal modelling these correspond to the underlying mechanisms, in engineering to the multiple fidelities present, and in gene-regulatory networks to the transcription factors. These functions are partially observed across the DAG, with noisy and heterogeneously sampled measurements, posing significant challenges for reconstruction, uncertainty propagation, and inference. To tackle these we place priors over functions and naturally arrive at Deep Gaussian Processes over DAGs. We theoretically study their prior-collapse behaviour, and the effect of graph topology and intermediate observations on the preservation of information. We obtain almost-sure lower bounds on the asymptotic frequency of depths at which distinction between inputs is preserved, identify broad kernel classes for which these hold, and prove an observation by \citep{dunlop2018} on the role of input connections. We offer a structured variational approximation that retains graph dependencies, satisfies compositional uncertainty, and captures the explaining-away behaviour of colliders. Finally, we empirically validate our theoretical results and our methodology, and model a latent-collider DAG, a protein signalling network, and a multi-fidelity heavy-ion collision task, attaining state-of-the-art performance while recovering the low-fidelity contributions and yielding explainability over the simulator hierarchy.

Open-ended reasoning and long-form generation tasks lack reliable automatic verification signals for reward-based policy optimization. Rubrics offer a promising alternative, but existing approaches treat them as given artifacts—either hand-crafted or prompt-generated—and often miss the task-specific, knowledge-intensive dimensions that matter most, distorting the reward signal. Our key observation is that rubric construction is itself a research problem: identifying what makes a response correct or insightful requires discovering and synthesizing external knowledge. We propose Deep Research as Rubric (DR-rubric-8B), a two-stage framework for constructing such rubrics. Stage I elicits domain facts, structural constraints, and failure modes through iterative multi-turn agentic search; Stage II distills this evidence into atomic, independently verifiable constraints for GRPO-based policy optimization. Because the model under training can serve as its own rubric generator, DR-rubric-8B supports bootstrap rubric generation without frontier-model assistance. We evaluate on 6 benchmarks spanning agentic research and expert reasoning. Experiments show that DR-Rubric achieves strong competitive performance with only 1K–3K training instances, where GPT-5-generated rubrics particularly benefit breadth coverage on agentic tasks, while bootstrap rubrics exhibit a specialization-to-rebalancing evolution and achieve the best overall performance at the third iteration. Results demonstrate that reframing rubric construction from static evaluation templates into an evidence-driven research process yields more scalable, fine-grained reward signals for open-ended tasks.

Skill-augmented agents load reusable skills as persistent runtime context, improving task performance but also giving malicious skills a durable channel for steering future actions. Such skills may leak secrets, corrupt code, bypass approvals, or stage data for exfiltration only after a concrete user task and workspace state make the unsafe action appear useful. This makes pre-install vetting insufficient and calls for runtime, task-conditioned protection. We propose Defense-as-Skill, a defense paradigm that implements the runtime guard itself as an installable, inspectable, and editable skill. Our guard, SkillSonar, runs alongside untrusted task skills and checks sensitive actions against the user's task boundary, routing each action to an allow, replan, or confirmation decision without modifying the underlying agent runtime. To study this setting, we construct SCOPE-R, a task-conditioned dataset covering 6 risk families and 21 sub-categories, with 206 attack-confirmed malicious instances and 43 benign tasks. We then improve SkillSonar on the SCOPE-R training subset using runtime guard-skill evolution, a Monte-Carlo Tree Search procedure that evolves the on-disk guard skill from feedback on the rollouts. On held-out SCOPE-R test splits across Claude Code and OpenClaw backends, SkillSonar reduces attack success while preserving benign utility, maintaining modest token overhead, and transferring to OOD attack families. These results show that runtime guards can be skill-native, transparent, and evolvable policy artifacts for skill-augmented agents.

We introduce ACMA (Delay-Augmented Clustering and Model Approximation), a noise-robust pipeline for saccade detection in oculographic signals that couples delay-embedded clustering with physiologically grounded parametric refinement. Existing detectors face a recurring trade-off: velocity-thresholding methods are simple and interpretable but degrade catastrophically as recording noise grows, adaptive event-based detectors recover part of this gap at moderate noise but lose reliability at high noise, and recent deep models are sensitive to noise distributions unseen during training. ACMA addresses this trade-off in two stages. First, a sliding-window two-cluster decomposition with time-delay embedding identifies candidate saccades while remaining stable across noise regimes. Second, a parametric saccade waveform constrained by priors on amplitude, duration, and the saccadic main sequence validates and characterizes each event. We evaluate ACMA on a saccade-detection benchmark spanning a wide range of noise levels and nine representative baselines, including a leading deep-learning detector (U'n'Eye), adaptive methods (NH, REMoDNaV, Engbert), clustering-based I2MC, and classical velocity-thresholding methods (IVT, IVVT, IDT, IDVT). ACMA delivers consistent F1 gains in moderate-to-high noise regimes, where competing detectors degrade sharply, while additionally returning physically meaningful per-event metrics (amplitude, duration, peak velocity) for downstream neurological assessment, fatigue monitoring, and brain-computer interface applications.

While label-conditional diffusion models exhibit remarkable generative capabilities recently, their success heavily relies on massive, cleanly labeled datasets. In practice, categorical supervision is rarely perfect: it is often corrupted by noise, clouded by ambiguity, or partially missing. Training directly on such weak annotations severely degrades generation quality, and existing robust methods offer only fragmented solutions that depend on scarce auxiliary priors. To overcome these limitations, we propose DELTA, a unified framework that robustly trains diffusion models across all three weak annotation types without any external priors. By treating the unknown true label as a latent variable, DELTA optimizes a principled variational objective that jointly recovers the clean data distribution and infers the true label posteriors. To make this joint training computationally tractable, we further introduce a median-centered timestep sampling strategy that efficiently concentrates evaluation where the diffusion process is most informative. Extensive experiments demonstrate that DELTA produces high-fidelity, class-consistent samples across diverse weak supervision scenarios, outperforming specialized baselines while demanding strictly less prior knowledge.

Muon has recently demonstrated strong empirical performance in large language model training, but the theoretical role of momentum in Muon remains unclear. Existing analyses of Muon either remove momentum to study spectral updates in isolation, or retain momentum without explaining why it improves empirical performance. Our work bridges this gap by showing momentum in Muon acts as a spectral filter. Under a structured signal-plus-perturbation gradient model, we prove that momentum suppresses perturbations while preserving the dominant signal, thereby enlarging the spectral gap between them. This enlarged gap stabilizes the singular subspaces of the matrix passed to Muon's orthogonalization step, making the resulting update more reliable. We further show that applying momentum before orthogonalization achieves provably stronger alignment with the signal component of the gradient than either reversing this order or simply removing momentum. Experiments across diverse tasks, including LLM training, support our theoretical analysis. More broadly, our theory offers a starting point for understanding the benefits of momentum in other matrix-based optimizers.

Discrete diffusion language models (dLLMs) accelerate text generation by unmasking multiple tokens in parallel. However, parallel decoding introduces a distributional mismatch: it approximates the joint conditional using a fully factorized product of per-token marginals, which degrades output quality when selected tokens are strongly dependent. We propose DEMASK (DEpendency-guided unMASKing), a lightweight dependency predictor that attaches to the final hidden states of a dLLM. In a single forward pass, it estimates pairwise conditional influences between masked positions. Using these predictions, a greedy selection algorithm identifies positions with bounded cumulative dependency for simultaneous unmasking. Under a sub-additivity assumption, we prove this bounds the total variation distance between our parallel sampling and the model's joint. Empirically, DEMASK achieves 1.7--2.2$\times$ speedup on Dream-7B while matching or improving accuracy compared to confidence-based and KL-based baselines. When applied to dParallel, a block diffusion model, DEMASK also produces Pareto-optimal accuracy-step trade-offs across math and code benchmarks.

Depth-recurrence promises improved latent reasoning by sharing parameters across depths. However, prior work still relies on partially fixed layer stacks and overlooks the bottleneck of constant hidden-size; moreover, it largely lacks rigorously resource-matched ablations. To address this, we introduce depth-recurrent attention mixtures (Dreamer), a modular architecture with a single recurring layer. Concretely, we mix sequence attention, depth attention, and sparse expert attention. This alleviates the hidden-size bottleneck, decouples scaling dimensions, and achieves effective yet efficient depth-recurrence. Across natural language reasoning benchmarks at up to ~2B parameters, Dreamer requires on average ~3x fewer training tokens for the same accuracy as FLOP-, parameter-, and memory-matched modern Transformers, and outperforms ~2x larger Transformers given same training tokens. Analysis further reveals 2-11x greater expert selection diversity than conventional MoEs, highlighting the flexible knowledge sharing across depths.

Hyperparameter transfer has become an important component of modern large-scale training recipes. Existing methods, such as muP, primarily focus on transfer between model sizes, with transfer across batch sizes and training horizons often relying on empirical scaling rules informed by insights from timescale preservation, quadratic proxies, and continuous-time approximations. We study hyperparameter scaling laws for modern first-order optimizers through the lens of recent convergence bounds for methods based on the Linear Minimization Oracle (LMO), a framework that includes normalized SGD, signSGD (approximating Adam), and Muon. Treating bounds in recent literature as a proxy and minimizing them across different tuning regimes yields closed-form power-law schedules for learning rate, momentum, and batch size as functions of the iteration or token budget. Our analysis, holding model size fixed, recovers most insights and observations from the literature under a unified and principled perspective, with clear directions open for future research. Our results draw particular attention to the interaction between momentum and batch-size scaling, suggesting that optimal performance may be achieved with several scaling strategies.

Designing regulatory DNA with cell-type-specific activity is broadly relevant for cell engineering and gene therapy. The regulatory activity of DNA elements such as promoters and enhancers emerges from the combinatorial organization of transcription factor binding sites and sequence context, which collectively constitute the genome's regulatory grammar. Deep learning models can effectively predict DNA regulatory activity across cell types, but current generative approaches often produce DNA sequences that strongly deviate from this natural regulatory grammar, increasing the risk of unexpected behavior and off-target effects in real-world applications. Here, we introduce DNA-CRAFT, a method for designing regulatory DNA with high predicted cell-type-specific activity while preserving naturalness. DNA-CRAFT formulates regulatory sequence design as a guided search over a generative model pre-trained on millions of naturally occurring regulatory elements from the human and mouse genomes. Our method enables targeted exploration of DNA sequence space, efficient evaluation of candidate designs, and explicit suppression of unwanted activity. Across benchmarks spanning human cell lines and primary immune cells, DNA-CRAFT outperforms existing generative and optimization-based methods, consistently achieving the best trade-off between cell-type specificity and naturalness in the designed regulatory DNA sequences.

Diffusion-based generative AI has achieved remarkable success in e-commerce applications such as virtual try-on, poster generation, and product background synthesis. However, when making online purchasing decisions for apparel, consumers also desire the freedom to examine specific detail regions of interest, such as collars, cuffs, and fabric textures, yet existing methods have not explicitly studied this setting. We therefore formalize a new, non-template task: Fashion Detail Generation with focus conditioning, and release FDBench, the first dataset and benchmark comprising 40K+ human-verified reference-detail pairs across 41 different categories. This task poses a unique semantic gap challenge: the model must bridge the correspondence between a focus marker on a product reference image and a photorealistic close-up view of the indicated region, while faithfully preserving the garment's identity, without any precise prompt. To bridge this gap, we propose Cross-modal Feature Alignment Distillation (CFAD), which leverages a fine-tuned DINOv3 teacher to align both branches of a Multimodal Diffusion Transformer in a shared semantic space via dual-branch distillation. To further improve consistency between generated details and reference images, we introduce a consistency reward model that jointly scores image pairs along three quality axes and optimizes generation via reinforcement learning. Experiments show that our model DetailAnywhere significantly outperforms all state-of-the-art opensource methods across all metrics and human evaluations.

Hardware-enabled monitoring of GPU workloads underpins many proposals for AI compute governance, but if developers can defeat monitoring mechanisms, such schemes are unworkable. We evaluate the adversarial robustness of GPU workload classification using only zero-overhead, privacy-preserving NVML telemetry: content-agnostic signals that observe physical effects of computation without accessing model weights, training data, or hyperparameters. Across 5 rounds of monitor-evader iteration, we evaluate 20 evasion strategy families on 9 GPU models spanning 4 architecture generations. We develop a classifier that achieves 98.6% binary accuracy at identifying training workloads across the whole corpus, and 43-87% accuracy against the most challenging unexpected workloads even when they are adversarially disguised.

The prevailing paradigm in large language model (LLM) alignment operates via erasure, filtering unsafe data or training models to strictly refuse harmful prompts. While effective at reducing immediate toxicity, this approach fundamentally constricts the model's epistemological scope, resulting in over-cautious systems that output uninformative blanket refusals to sensitive yet benign queries. In this work, we challenge the orthodoxy that unsafe data must be discarded. We propose a dialectical approach to alignment, positing that unsafe data encodes rich, domain-specific knowledge critical for nuanced, safe, and informative generation. To operationalize this, we introduce SafeMoE, a Mixture-of-Experts (MoE) framework that isolates unsafe knowledge into domain-specific Low-Rank Adapters (LoRA experts) trained exclusively on harmful corpora. To synthesize safety from these unsafe primitives, we train a lightweight gating network using a minimal, highly curated set of safe-informative responses (as few as 100 samples). During inference, this router dynamically orchestrates the unsafe experts, effectively steering the generation trajectory to harness their deep domain knowledge while strictly enforcing safety constraints. Extensive empirical evaluations across stringent safety benchmarks demonstrate that SafeMoE is not only safer, achieving over a 20\% relative improvement in safe response rate (more than a 15\% absolute gain), but also produces more informative responses when safety and harmfulness are of paramount concern. Furthermore, the routing mechanism exhibits strong zero-shot generalization to unseen domains and broader safety tasks without domain-specific supervision. Our findings suggest a paradigm shift in alignment: true safety requires not the masking of unsafe knowledge, but its controlled integration.

More than 80% of the 1.6B English speakers do not use Standard American English (SAE), yet LLMs often fail to correctly identify non-SAE dialects and generate stereotyped responses for their speakers. We introduce DialectLLM, the first large-scale framework for generating high-quality multi-dialectal conversational data encompassing the three pillars of written dialect---lexical (vocabulary), orthographic (spelling), and morphosyntactic (grammar) features. DialectLLM produces a dialect-parallel dialog dataset spanning nine English dialects. Partnering with native linguists, we design and validate SAE-to-dialect transformation rules, ensuring authenticity. Our approach challenges the prevailing practice of applying a single morphosyntactic feature set to both user utterances and model responses, showing that models should not reproduce up to 90% of the grammatical features of a dialect. Human evaluation confirms data quality, with annotators preferring DialectLLM over prior methods in 98.8% of pairwise comparisons for dialect naturalness. We then construct DialectLLM-Bench, a dialect-parallel benchmark with 50k+ dialogs, resulting in 97k+ QA pairs, and evaluate 17 LLMs on dialect identification and response generation tasks. Even frontier models achieve under 70% accuracy, fail to reach 50% for prominent dialects like Canadian English, and systematically misclassify non-SAE dialects as American or British. Beyond benchmarking, we show that DialectLLM data also serve as a scalable LLM post-training resource, suggesting a practical path toward dialect-aware conversational AI.

Recent advances in semiconductor technology have dramatically improved IC chip performance, but soaring transistor densities and clock speeds generate unprecedented heat fluxes in modern integrated circuits, making spatially adaptive thermal management critical. However, conventional heat sinks rely on fixed geometries that lack customization for chip-specific thermal profiles, while the absence of optimal design datasets and strict manufacturing constraints hinder the application of generative AI. To address these challenges, we present DiffCool, a physics-guided framework that reformulates heat sink synthesis as a constrained discrete denoising process. By embedding manufacturing-aware state transitions into the diffusion dynamics and optimizing via a differentiable thermal-aware loss coupled with a surrogate simulator, our model generates structurally sound, high-performance topologies directly from bare-chip heat maps without labeled data. Experiments on an open-source processor demonstrate that DiffCool synthesizes production-ready designs in under one second, reducing peak temperature by 28.3% and temperature gradient by 56.9% compared to conventional baselines. This work establishes a scalable, label-free paradigm that unifies generative modeling with physical constraints for automated thermal design.

State estimation is a critical challenge in modeling dynamic systems. Traditional Kalman filters and data-driven methods have shown good results in complex, nonlinear systems. However, most existing methods optimize the posterior without considering prior modeling, which leaves room for improvement in state estimation accuracy, and predict high-dimensional Kalman parameters end-to-end, which lead to unstable training and poor convergence. To address this, we propose Diff-Kalman, which leverages state differences to predict state increments and scaling matrices for correcting both the prediction and update phases of the Kalman filter. Diff-Kalman combines traditional dynamic models with neural networks to adaptively correct system behavior. In the prediction phase, Diff-Kalman utilizes a Transformer-based architecture with self-attention to extract localized patterns from historical states, followed by cross-attention to predict future state increments. These increments are then fused with model-based priors from the dynamic model. In the update phase, a GRU-based Scaling Predictor (GSP) dynamically adjusts the Kalman covariance and gain matrices based on differential residuals, which improves the accuracy and robustness of the estimation. Experimental results on linear MOT trajectories and nonlinear Lorenz dynamics show that Diff-Kalman improves estimation accuracy over representative baselines.

Autoregressive language models are widely used for text evaluation, however, their left-to-right factorization introduces positional bias, i.e., early tokens are scored with only leftward context, conflating architectural asymmetry with true text quality. We propose masked reconstruction as an alternative paradigm, where every token is scored using full bidirectional context. We introduce DiffScore, an evaluation framework built on Masked Large Diffusion Language Models. By measuring text recoverability across continuous masking rates, DiffScore eliminates positional bias and naturally establishes an evaluation hierarchy from local fluency to global coherence. We further provide diagnostic tools unavailable to autoregressive frameworks: multi-timestep quality profiles that decompose scores across masking rates, and bidirectional Pointwise Mutual Information (PMI) decomposition that disentangles fluency from faithfulness. Experiments across ten benchmarks show that DiffScore consistently outperforms autoregressive baselines in both zero-shot and fine-tuned settings.

Offline goal-conditioned reinforcement learning (GCRL) learns goal-directed policies from reward-free data, but in long-horizon tasks, goal-conditioned value functions often provide unstable guidance due to sparse rewards and discounting. Hierarchical methods partially mitigate this issue via subgoal decomposition; however, high-level decision-making still relies on noise-sensitive value estimates, leading to unstable behavior in complex environments. We address this limitation by proposing \textbf{D}iffusion \textbf{S}ubgoal \textbf{P}lanning (\textbf{DSP}), a diffusion-based framework for high-level subgoal generation. DSP casts high-level planning as guided generative inference over goal-conditioned subgoals and learns both conditional and unconditional flows, enabling classifier-free guidance to introduce a goal-directed bias at inference time. By removing explicit value-based guidance from high-level planning, DSP generates reachable and goal-directed subgoals through a generative model while retaining hierarchical execution. Experiments on offline GCRL benchmarks demonstrate that DSP outperforms prior methods on a range of navigation and manipulation tasks, with particularly strong performance in maze environments that require multi-step subgoal planning.

Reconstructing audio from brain activity remains challenging. A key limitation lies in current audio diffusion models, which are primarily designed for text-based conditioning and lack mechanisms to incorporate non-linguistic continuous signals such as neural responses. While recent work in the visual domain has explored direct conditioning on brain activity, this approach has not yet been extended to audio diffusion models. We introduce a framework that directly guides a latent audio diffusion model with fMRI signals, avoiding intermediate feature prediction for the conditioning stage. Central to our approach is a transformer-compatible adapter that enables the integration of non-linguistic representations into the cross-attention layers of diffusion transformers (DiTs). Although motivated by neural decoding, this adapter provides a general mechanism for conditioning audio diffusion models on arbitrary continuous inputs, not inherently limited to neural signals. We demonstrate consistent improvements over baselines, including two-stage fMRI-to-latent conditioning, across both acoustic and semantic evaluation metrics. In addition, we introduce a fidelity metric grounded in a computational model of auditory neural processing, which allows us to quantify the contribution of individual auditory cortical regions to reconstructed audio representations.

Generating high-quality UV textures for 3D meshes remains challenging. Multi-view projection pipelines suffer from occlusion and view inconsistency, and recent methods that generate textures directly in UV space still rely on auxiliary modules to supply 3D information, leaving the attention mechanism tied to UV-grid positions rather than to the underlying surface geometry. This mismatch limits coherence across seams and disconnected UV islands. We propose DirectUV, an image-conditioned UV texture diffusion framework that operates in the latent UV space of a pretrained image VAE, in which a Diffusion Transformer denoises the UV latent given a single input image and a coarse UV map. At its core, Surface-Aware Positional Encoding (SAPE) replaces the standard 2D-grid positional encoding with encodings derived from per-token 3D surface coordinates obtained via UV-to-surface correspondence. As positional encodings define the distance metric that attention operates on, SAPE enables tokens to attend to each other based on true surface proximity rather than UV-grid distance, restoring coherence across seams and disconnected islands. A multi-level extension further assigns different attention heads to progressively finer subdivisions of the same latent UV patch, allowing the model to reason about surface structure at multiple granularities. Experiments show that DirectUV produces sharper and more globally consistent textures than other baselines, with the largest improvements in occluded and view-unseen regions where projection-based methods leave gaps or stretched textures.

Recent advances in reinforcement learning have made verifiable rewards a central post-training paradigm for improving multimodal reasoning. However, methods such as GRPO remain data- and compute-inefficient, often spending rollout and annotation budget on prompts that provide little learning signal. We propose DISCOVER, an online data-discovery framework for budgeted multimodal GRPO that enables vision-language models to dynamically identify prompts that are most useful for the current policy. DISCOVER first organizes the multimodal sample pool into semantic clusters and queries a small set of density-diverse representative samples to obtain initial reward observations. During training, it estimates the utility of unannotated samples by transferring sparse reward feedback from queried samples. Since sample utility changes as the policy evolves, DISCOVER dynamically updates these estimates with recency-weighted rollout observations. It then selects only high-utility prompts for annotation using a policy-conditioned score that accounts for reward variance and difficulty, and continues utility-based replay once the discovery budget is exhausted. Experiments on ViRL39K and VLAA-Thinking with Qwen3-VL-2B show that DISCOVER substantially improves multimodal mathematical reasoning under a strict 10\% unique-sample budget, outperforming budget-matched random selection, hard-example sampling, full-pool GRPO under the same training-step budget, and online selection baselines. Ablations further validate the importance of semantic coverage, online reward transfer, and stable discovery cadence, establishing DISCOVER as a practical framework for data-efficient multimodal RL post-training.

Frontier LLMs now perform strongly across a wide range of physics evaluations, but it is hard to disentangle genuine reasoning from recall of established science. We introduce DiscoverPhysics, an interactive benchmark that asks a LLM agent to discover the laws of motion of a simulated world whose physics deliberately deviates from our own. We construct eleven worlds governed by, among others, screened and fractional-power gravity, multi-species couplings, hidden dark-matter-like particles, non-coordinate-free physics, and time-varying interactions. Each world is generated on demand by an N-body simulator, for which the agent proposes several rounds of experiments, observes raw trajectory data, and ultimately submits both a natural-language explanation of the world's physics and a Python implementation of the inferred law. Because solving a world requires the agent to design informative experiments and revise its hypotheses, the benchmark probes long-horizon reasoning over an experimental history. We evaluate submissions along two complementary axes: trajectory MSE on held-out particles and an LLM-judged explanation score following an expert-written rubric assessing conceptual understanding of each world. Across ten frontier models, we find that the strongest agents pass only about half of the worlds and consistently fail on those where latent structure must be uncovered. Open-source models lag substantially behind commercial models, both in their ability to design informative experiments and in extracting conclusions from the data. We further find that good predictive accuracy does not guarantee high explanation quality and that conceptual understanding depends on hypothesis refinement through well-chosen experiments.

Biological circuits learn without catastrophic forgetting, but the structural basis of this ability remains unclear. We investigate whether distance-dependent connectivity (DDC), spatial recurrency found in the mammalian cortex, contributes to continual learning by embedding neurons of a biologically plausible recurrent spiking neural network in a 3D Euclidean substrate, using pairwise distance to set recurrent connection probability, and ablating that distance dependence in silico. We show that DDC shapes a synaptic resource geography: it controls the spatial breadth of the candidate synaptic pool for learning to occur, plasticity further compresses that pool, and the resulting substrate governs how subsequent inputs compete for the same synapses. On challenging 10-way class-incremental MNIST classification, this produces a non-monotonic accuracy curve, with an intermediate DDC range achieving the best performance (71.4%) and outperforming both highly local circuits that collapse into a synaptic-resource bottleneck and random connectivity that produces weakly guided, broader synaptic contention. This pattern was better explained by synaptic competition, resulting in separable neural dynamics, than by neuron assembly separation. These results suggest that appropriate cortical DDC may prevent diffuse random contention. For continual learning problems, this finding highlights the conceptual importance of circuit dynamics more than active neuron ensembles and engrams.

The canonical orientation of a 3D object is often not unique but inherently ambiguous: many objects admit multiple equally valid canonical frames that no single-output regressor can faithfully represent. Rotational symmetry is the most structured manifestation of this phenomenon, where the set of valid orientations forms an orbit of the symmetry group. Existing methods fall short in one of three ways: they predict only a single orientation, handle symmetry only under pre-defined group assumptions, or resort to test-time augmentation to approximate the full set of valid orientations. We instead introduce a general framework for ambiguity-aware orientation estimation that is at once $SO(3)$-equivariant by construction, distributional in its output, and free of any restriction on the symmetry group. It effectively predicts multiple valid orientations as a continuous multi-modal distribution over $SO(3)$ via a truncated Wigner-D expansion, without any symmetry-group assumption. The proposed method outperforms recent equivariant regressors and fixed-quotient classifiers on the ShapeNet orientation benchmark, particularly on categories with continuous rotational symmetry, showing that supervision on discrete symmetry labels generalises to continuous bands.

Domain randomization (DR) improves sim-to-real transfer by training policies over randomized simulator dynamics, but standard DR optimizes average performance under a fixed sampler and can miss rare failure-prone domains. We propose Risk-Sensitive Domain Randomization (RSDR), a distributionally robust formulation of episodic DR that evaluates each policy against the worst-case distribution over dynamics parameters within a KL neighborhood of a reference sampler. The resulting soft adversary has an exponential-tilting form, yielding a single temperature-controlled sampler family: negative temperatures emphasize low-return domains for robustness, zero recovers uniform DR, and positive temperatures give an optimistic sampler closely related to curriculum-style DR. Because this target changes with the policy and returns are observed only through rollouts, RSDR learns an amortized dynamics sampler by reverse-KL variational inference while reusing on-policy PPO trajectories. Across six domain-randomized MuJoCo Playground tasks, robust RSDR improves CVaR and minimum-return metrics over uniform and adaptive DR baselines, and outperforms best-tuned EPOpt.

Existing robust preference optimization for language-model alignment mainly studies pairwise supervision and places robustness at the dataset, prompt, or preference-pair level. We instead study listwise preference optimization under ranking-label uncertainty: given a prompt and a candidate list, the observed ranking over that list may be ambiguous due to annotator inconsistency, near-ties, lossy rankwise feedback, or reward-model noise. We propose a pointwise total-variation robust Plackett--Luce objective that directly robustifies the ranking label conditional on the candidate list. The robust loss admits an exact decomposition into the nominal PL loss plus a worst-case PL correction, and the worst-case ranking is obtained by sorting current implicit scores in ascending order, reducing the inner maximization from $K!$ enumeration to $O(K\log K)$. This tractable structure yields strong offline and online optimization guarantees. In the offline fixed-list setting, the robust objective is convex and projected stochastic subgradient reaches global $\epsilon$-suboptimality with $O(\epsilon^{-2})$ sample complexity. In the online policy-induced setting, where candidate lists are generated by the current policy, we establish weak convexity and $\widetilde O(\epsilon^{-2})$ Moreau-envelope stationarity. Experiments in offline LLM alignment show that the proposed robust correction largely preserves performance under clean labels and improves robustness under noise. In online alignment, it makes reward-model-ranked candidate expansion more reliable and improves both reward-model andexternal GPT-4 judge metrics.

Missing data imputation, where a model is trained on observed data to estimate unobserved values, is a fundamental problem in machine learning. In this paper, we rigorously formulate imputation model learning as a mean-squared error risk minimisation problem. We show that when the probability of missingness depends on the data, many state-of-the-art methods fail to account for the resulting distribution shift between the observed data used for training and the full data distribution used for evaluation. Consequently, these approaches do not minimise mean-squared error on the full data distribution. Instead, we propose a novel imputation algorithm designed to learn an imputation model from the observed data while explicitly accounting for this distribution shift. Simulation studies show consistent improvements over otherwise identical uncorrected baselines, with average reductions of 3\% in RMSE and 7\% in Wasserstein distance.

Vision-Language-Action (VLA) models inherit their visual backbone from static image-text pretraining, leaving physical dynamics to be learned from scarce action data. Generative video models, by contrast, already encode motion, contact, and implicit physics at internet scale. We introduce DiT4DiT, a Video-Action Model (VAM) that couples a Video Diffusion Transformer with an Action Diffusion Transformer and co-trains them under a dual flow-matching objective with a tri-timestep scheme for video supervision, action supervision, and feature extraction. The key insight is that the policy does not need to generate a future video: we intercept the Video DiT's hidden state at a single fixed flow timestep, turning a multi-step video rollout into a one-shot feature extraction consumed by the Action DiT. Across benchmarks, DiT4DiT sets new SOTA averages on LIBERO (98.6%) and the 24-task RoboCasa-GR1 suite (56.7%), outperforming strong VLA and video-based baselines. On a real Unitree G1, DiT4DiT reaches 73.8% on eight tabletop tasks and 72.2% on three whole-body loco-manipulation tasks, running up to 12$\times$ faster than prior VAM baselines while also exceeding them in accuracy. To our knowledge, this is the first video-generation-based policy to run humanoid whole-body control at real-time rates. It further generalizes zero-shot to unseen object categories and scene variations. Together, these results indicate that video-dynamics priors, accessed through a single forward pass, are a practical and scalable foundation for generalist robot policies.

The advent of Multimodal Large Language Models (MLLMs) has unlocked the potential for end-to-end document parsing and translation. However, prevailing benchmarks such as OmniDocBench and DITrans are dominated by pristine scanned or digital-born documents. They do not adequately represent the intricate challenges of real-world capture conditions, such as geometric distortions and photometric variations. To fill this gap, we introduce DocPTBench, a comprehensive benchmark specifically designed for Photographed Document Parsing and Translation. DocPTBench comprises over 1,300 high-resolution photographed documents from multiple domains, includes eight translation scenarios, and provides meticulously human-verified annotations for both parsing and translation. Our experiments demonstrate that transitioning from digital-born to photographed documents results in a substantial performance decline: popular MLLMs exhibit an average accuracy drop of 18\% in end-to-end parsing and 12\% in translation, while specialized document parsing models show a more prominent average decrease of 22\%. This substantial performance gap highlights the unique challenges posed by documents captured in real-world conditions and reveals the limited robustness of existing models. The benchmark and related code will be made publicly available for further research.

A central question in modern learning theory is what notion of compression, if any, explains why over-parameterized neural networks generalize. Existing answers measure complexity at different levels, ranging from weights to representations, and estimating the associated quantities either requires held-out data in principle or relies on mutual-information estimation in ultra-high dimensions. We approach the question through the minimum description length (MDL) principle and propose the prequential regret as a complementary signal that is computable from the training trajectory alone. We establish that, in the (nearly) interpolating regime, the expected prequential regret upper-bounds the description length of the trained parameter under the algorithm-induced distribution, and that this description length translates into a high-probability generalization bound in the PAC-MDL setting. Along the way, we show that the encoder-complexity correction recently added to the algorithm-level information bottleneck is upper-bounded by an MDL parameter description length, bringing the two views of compression onto a shared footing. Empirically, the prequential regret tracks the generalization error on CIFAR-10 and CIFAR-100, with Pearson correlations exceeding 0.9.

Successful deep neural networks discover salient features of data. We show when and why they fail to learn out-of-distribution (OOD)-relevant representations from an in-distribution (ID) training window. This requires decoupling feature learning from data-generating-process (DGP) identifiability. From a single training window, OOD extrapolation is non-identifiable: infinitely many DGPs are $\varepsilon$-observationally equivalent on the training data but diverge arbitrarily outside it, and no in-distribution criterion alone reliably breaks the tie. A structural commitment, the feature map, label map, and model class $(\varphi, \psi, \mathcal{M})$, dictates the assumed DGP and governs OOD generalization while leaving ID performance essentially unchanged. When architecture, pretraining, augmentation, input formats, or domain knowledge implicitly inject the missing commitment, the model succeeds. When it cannot infer OOD-relevant structure from ID evidence, it fails. Changing only the representation can make the same architecture, at the same in-distribution loss, differ by ${\sim}520\times$ out of distribution. When the commitment is correct {\em and} identifiable, OOD error vanishes. For example, Fourier coordinates turn periodic extrapolation into interpolation on $\mathbb{S}^1$. The same mechanism predicts outcomes in three natural-science settings (mass-action chemistry; Kepler's-third-law exoplanet prediction, $n=2{,}362$; and cross-species coding-DNA detection) and in a 264-run positional-encoding study across Transformer, Mamba, and S4D. Finally, a controlled study shows: correct features are necessary but not sufficient. The model class must express the target, and the transformed training data must cover the relevant representation space. Thus, feature engineering is not a departure from what makes deep learning successful. It is the explicit structural commitment that makes extrapolation identifiable.

Standard embodied evaluations do not independently score whether an agent correctly commits to task completion at episode closure—a capacity we call terminal commitment. Behaviorally distinct failures—never completing the task, completing it but failing to stop, and reporting success without sufficient evidence—collapse into the same benchmark failure. We introduce VIGIL, an evaluation framework that makes terminal commitment independently measurable. Under VIGIL’s default protocol, agents observe only egocentric RGB, receive no action-success signals, and must end each episode with a semantic report checked deterministically against hidden world state. This yields two separate scores: world-state completion (W) and benchmark success (B), where B additionally requires a correct terminal report. This decoupling makes four outcome categories distinguishable: missed execution, post-attainment drift, unsupported commitment, and verified success. Across 20 models on 1,000 frozen episodes, systems with comparable W differ by up to 19.7 pp in B: one model converts achieved states into correct reports, while another with near-identical execution drifts past the goal without closing. An action-feedback intervention further tests the separation: execution-oriented signals improve W broadly, yet commitment failures persist in models that do not already ground terminal reports in the achieved state. VIGIL provides a protocol that makes terminal commitment independently visible and scorable.

RL post-training with methods like PPO and GRPO produce a large rollout archive, which each contain correctness-labeled completions, correct/incorrect completions for the same prompt, teacher log-probabilities, and group pass rates. Canonical distillation methods, however, ignore these traces and train based upon new data from the converged teacher. This discards the contrastive signal that helped the teacher and the metadata already produced during teacher RL, even though the teacher RL run typically generates many more tokens than the later distillation pass. However, reusing rollouts naively is non-trivial: the archive mixes early and late teacher policies, can be arbitrarily off-policy for the student, and contains many low-quality traces, and naive SFT and DPO on these rollouts underperforms the cannonical SFT baseline. In this paper, we show that rollout reuse becomes effective when it is treated as a data-selection problem. We use two sets of information from rollouts: per-prompt pass rate, which defines an easy-to-hard curriculum and separates all-correct prompts for SFT from mixed-success prompts for DPO; and per-rollout teacher likelihood, which acts as an in-distribution proxy for student-compatible traces. Across two GRPO teachers (4B and 8B) and students from multiple families, the curated rollout datasets DSFT^{allp,top1} and DDPO^{curr,top1} match or beat canonical distillation without an extra teacher generation pass. On MATH500, DDPO^{curr,top1} improves over Dsynth by up to 4.0 percentage points on strong students, while DSFT^{allp,top1} → DDPO^{curr,top1} matches weaker students where DPO is unstable.

Modern GPU-native solvers for combinatorial optimization exploit parallelism by evolving large populations of relaxed candidates; yet, their post-processing collapses this computational effort into a single, rounded solution. We show that the discarded replicas of Parallel Quasi-Quantum Annealing (PQQA) form an implicit elite archive that contains basin-level information recoverable after annealing. We introduce Iterative Path-Relinking Polish (I-PRP), a deterministic, training-free post-anneal operator that leaves the PQQA inner loop unchanged. I-PRP polishes the top rounded elites, traverses cost-guided relinking paths from the current best elite toward the others, re-polishes intermediate states to cross basin boundaries, and iterates only under strict improvement. The operator never degrades the paired upstream polish, terminates after a finite number of updates, and reduces to the original polish when the population collapses to a single basin. Across QUBO and graph-coloring benchmarks, I-PRP preserves saturated cases while improving PQQA on rugged multi-basin instances, revealing that parallel annealing populations contain reusable search structures beyond their best replicas.

Sparse autoencoders (SAEs) have become an important tool for unsupervised concept discovery in large models. To make the resulting feature spaces more interpretable and manageable, recent approaches have begun imposing hierarchical structure, either explicitly or as an implicit effect of training constraints, yet rigorous comparison remains difficult. There are no agreed-upon requirements for what a meaningful feature hierarchy should satisfy, and evaluation has largely relied on qualitative illustrations with fragmented quantitative protocols. To address this, we derive a set of key requirements for generalization/specialization hierarchies in unsupervised concept discovery, drawing on semantic net and taxonomy research alongside recent SAE work, and use them to derive a concrete evaluation protocol. Applying this protocol to current SAE approaches trained on visual data, we find that while feature spaces generally provide a basis for sensible hierarchies, establishing good hierarchical structure remains challenging. In particular, feature absorption, both in its well-known hard form and in a continuous, soft form, systematically compromises hierarchy quality, pointing to a fundamental tension that future approaches will need to navigate.

Reproduction test generation, producing a failing-then-passing test that captures a reported bug, is a critical step in automated software engineering. Existing agentic methods treat this as a monolithic loop, despite the task inherently comprising two subtasks of distinct nature: diagnosing the root cause and writing a fail-to-pass test. Without explicit separation, the agent faces a compound objective with underspecified intermediate goals, leading to goal drift. We propose DPIAgent, a structured agentic framework built on three principles, Divide, Protocol, Isolate (DPI), that mitigates compound-objective ambiguity and goal drift: it Divides the task into single-objective phases of defect exploration and test generation; enforces a handoff Protocol that records the diagnosis and test plan, preventing context loss; and Isolates each phase's action space by tailoring the toolset to its task, preventing irrelevant tools from misleading execution. On SWT-Bench Verified, DPIAgent outperforms seven baselines across three backbone LLMs. With DPI alone it reaches 81.76% success rate on GPT-5, the highest reported among open-source methods, gaining up to 11.88 points over the strongest baseline on GPT-5-Mini; adding test selection further raises it to 86.17%. Our analysis shows that architectural structure and backbone capability are complementary axes rather than substitutes, demonstrating DPI's generalizability across model classes.

Multi-modal models require visual tokenizers that jointly capture high semantic density and fine-grained structural fidelity. Representation Autoencoders (RAEs) offer a promising direction by decoding images directly from a pre-trained semantic space using high-dimensional continuous tokens, bypassing the information bottleneck of VAEs and benefiting both visual understanding and generation. In this work, we aim to discretize these continuous vision representations to bridge the gap with language models — a non-trivial challenge, as existing quantization methods suffer from codebook collapse, failing to scale while preserving semantic coherence. We identify the root cause as **metric mismatch**: standard Euclidean codebook objectives are fundamentally misaligned with the anisotropic geometry of representation space, leading to codebook embeddings with high-variance magnitude scales and uneven angular distributions that hinder scalability. To address this, we propose **Hyper-Spherical Quantization (HSQ)** , which decouples semantic content from feature magnitude via angular routing, preventing code assignment from being dominated by scale rather than meaning. The resulting **discrete Representation Autoencoder (dRAE)** achieves high-fidelity reconstruction while preserving semantic integrity and supporting scalable discrete latent spaces. Extensive experiments demonstrate consistent performance gains as the vocabulary size scales to $131,072$, along with strong unified performance across both visual understanding and generation benchmarks.

Vision-language models (VLMs) have made steady progress as in-vehicle driving assistants, yet existing evaluations largely remain offline, focusing on perception and short-term reasoning over pre-recorded driving scenes. A complementary capability is still underexplored: whether a model can continuously monitor a causal driving video stream and issue a calibrated alert when a user-specified condition is met. Existing driving benchmarks mainly assess reactive understanding, while recent streaming video benchmarks are developed largely outside the driving domain. To this end, we introduce DriveStreamBench, a driving and streaming video benchmark that organizes this capability spectrum into four cognitive levels constructed from a shared pool of driving events, to support level-wise capability profiling and gated evaluation. We evaluate various vision-language models along with DriveStream-SFT, a supervised reference baseline trained on the benchmark's training split, under a unified streaming protocol. DriveStreamBench tests a basic but necessary capability, and all evaluated models fall short of it; current vision-language models are not yet ready for user-conditioned alerting in driving. Resources are available at https://anonymous.4open.science/r/DriveStreamBench/ and https://dataverse.harvard.edu/previewurl.xhtml?token=01f6bc4c-d809-4e40-b025-b6d783c86571.

Joint-Embedding Predictive Architectures (JEPAs) are prone to representation collapse, typically mitigated through empirical heuristics. We develop an early-training stability theory that unifies these heuristics. Linearising the coupled JEPA gradient flow around the trivial fixed point reveals two competing effects: a driving force ($\gamma$) and a decay effect ($\sigma$). Under approximate spectral decoupling, a per-mode stability ratio $\mu_i = \gamma_i / \sigma_i$ factorises into independent data-side and predictor-side terms and the count of unstable modes tracks the rank of representations that can emerge. The framework predicts a phase boundary, which we confirm empirically across more than 800 Tabular-JEPA configurations. It also unifies predictor scaling, masking ratio, and EMA as distinct mechanisms for shifting $\mu$. Guided by this analysis, we introduce ResidualPred, a transformer predictor whose attention is biased toward the identity at initialisation; it improves both effective rank and downstream accuracy on tabular benchmarks and on I-JEPA pretraining over CIFAR-10, CIFAR-100, and STL-10. Our framework connects empirical collapse-avoidance heuristics to an explicit dynamical picture, yielding theory-driven stabilizers. Code is available as supplementary material.

The default representation (DR), originally introduced in neuroscience, and its principal eigenvector have been shown to be effective for a wide variety of applications, including reward shaping, count-based exploration, option discovery, and transfer. However, in prior investigations, the eigenvectors of the DR were computed by first approximating the DR matrix, and then performing an eigendecomposition. This procedure is computationally expensive and does not scale to high-dimensional spaces. In this paper, we propose an objective for approximating the principal eigenvector of the DR using only transitions with a neural network. This objective is inspired by a series of theoretical results and is empirically validated in a number of environments. We then demonstrate its usefulness by applying the learned eigenvectors for reward shaping.

Video diffusion models (VDMs) have achieved impressive progress in text-to-video generation, but their high memory and computational costs hinder practical deployment. Quantization-aware training (QAT) is an effective solution for compressing and accelerating advanced generative models without runtime overhead at inference. % However, existing QAT methods suffer from a distinctive challenge in VDMs: while they often preserve prompt semantics, global layout, and coarse motion, the quantized model severely degrades visual details, texture fidelity, and sharpness. In this paper, we trace this degradation to the timestep-agnostic design of conventional quantization pipelines, which overlooks the stage-wise functionality of video denoising. In VDMs, early denoising steps mainly establish global structure and motion, whereas middle and late steps refine local appearance and high-frequency details. Based on this insight, we propose \textbf{DSAQuant}, a \textbf{D}enoising-\textbf{S}tage-\textbf{A}ligned \textbf{Quant}ization-aware training framework for VDMs. During training, \textit{Denoising-Stage Oriented Supervision} preserves teacher distillation in early steps for stable structure planning, while shifting later steps toward target-driven optimization to enhance detail reconstruction. During inference, \textit{Denoising-Stage Gated Guidance} disables CFG in the final denoising steps to prevent it from amplifying quantization-induced errors into high-frequency artifacts. Extensive experiments on the Wan and CogVideoX families under W4A4 and W3A3 settings show that DSAQuant consistently outperforms the SOTA QAT baseline, improving the VBench average score by up to 6.60 under aggressive W3A3 quantization while preserving strong text-video alignment. These results demonstrate that effective VDM quantization requires not only reducing quantization error, but also aligning quantization training and inference with the stage-wise nature of video diffusion.

Traffic flow prediction is a fundamental task in intelligent transportation systems. Existing methods primarily rely on temporal dependency modeling and thus face two critical limitations: struggling to effectively capture diverse periodicity; overlooking the relationship between spectral structures in the traffic flow and spatiotemporal traffic behaviors. To address these limitations, we propose DSSNet (Deep Spectral Structure Profiling Network), an adaptive spectral profiling framework. DSSNet follows the paradigm of "spectral structure profiling" and then "dual-domain modeling". Initially, a Spectral Structure Learning module captures the dominant frequency bands of the spectrum to enhance regional representations while simultaneously serving as discriminative spatiotemporal identity between regions. Based on these identities, we construct a global Resonance Graph to uncover cross-region dependencies and introduce a Dual-Domain Knowledge Integration module to aggregate structural knowledge across both time and frequency domains. Extensive experiments on multiple benchmark datasets demonstrate that DSSNet achieves state-of-the-art performance, while highlighting the potential of frequency domain modeling for understanding spatiotemporal traffic behaviors.

Architecture attribute predictors reduce the cost of neural architecture search by estimating accuracy or latency from a small set of evaluated candidates. The key difficulty is that neural architectures are operation-labeled DAGs: edge direction defines computational flow, but existing predictors often treat direction as a local structural cue. We propose DTA-GT, a Direction- and Topology-Aware Graph Transformer based on encoder-wide directional consistency, which preserves directed computational semantics across node initialization, global spectral encoding, pairwise interaction, and feature update. DTA-GT realizes this principle with direction- and topology-aware node initialization, Magnetic Laplacian spectral encoding, direction-dual structural attention, and a Direction-Sensitive MoE-FFN. Across accuracy and latency prediction on NAS-Bench-101 and NAS-Bench-201, DTA-GT consistently outperforms representative sequence-, GNN-, Transformer-, and hybrid predictors under limited-budget settings. Ablations and Magnetic spectral controls show that the improvements come from coordinated direction-aware stages and edge-orientation-sensitive spectral information, while mechanistic diagnostics confirm that the modules exhibit the intended directional behavior. These results support encoder-wide directional consistency as an effective design principle for architecture DAG representation.

Consistency-based generative models like Shortcut and MeanFlow achieve impressive results via a target-aware design for solving the Probability Flow ODE (PF-ODE). Typically, such methods introduce a target time $r$ alongside the current time $t$ to modulate outputs between a local multi-step derivative ($r = t$) and a global few-step integral ($r = 0$). However, the conventional "one input, one output" paradigm enforces a partition of the training budget, often allocating a significant portion (e.g., 75% in MeanFlow) solely to the multi-step objective for stability. This separation forces a trade-off: allocating sufficient samples to the multi-step objective leaves the few-step generation undertrained, which harms convergence and limits scalability.To this end, we propose Duality Models (DuMo) via a "one input, dual output" paradigm. Using a shared backbone with dual heads, DuMo simultaneously predicts velocity $\mathbf{v}_t$ and flow-map $\mathbf{u}_t$ from a single input $\mathbf{x}_t$. This applies geometric constraints from the multi-step objective to every sample, bounding the few-step estimation without separating training objectives, thereby significantly improving stability and efficiency. On ImageNet $256 \times 256$, a 679M Diffusion Transformer with SD-VAE achieves a state-of-the-art (SOTA) FID of 1.79 in just 2 steps.Code will be publicly available.

Large Vision Language Models (LVLMs) have demonstrated outstanding capabilities in multimodal content understanding, yet their multimodal nature also leads to security vulnerabilities. Existing defense methods often suffer from high computational cost and inference latency. In contrast, steering-based methods avoid these issues but still depend on heuristic model representations and struggle to handle coordinated multimodal attacks, where malicious content from images or text prompts can hijack the model’s limited attention resources. We identify this phenomenon as attention distraction, where an LVLM’s attention shifts from safety reasoning toward maliciously aligned tokens. To this end, we propose DualSteer, a lightweight and training-free defense framework that restores safe reasoning via dual-space steering. DualSteer first adjusts the attention distribution to counter the influence of jailbreak distractors, and then steers hidden layer representations to address residual jailbreak effects that the model cannot resist on its own. Extensive experiments across six jailbreak benchmarks and three LVLMs of different architectures demonstrate that DualSteer consistently reduces attack success rates by over 30\% relative to state-of-the-art defenses, while ensuring the model’s utility and avoiding false rejections. DualSteer provides a novel defensive perspective while offering an efficient and robust solution to defend against complex multimodal jailbreak attacks.

Diffusion and flow-matching based text-to-speech (TTS) models excel in naturalness but often lack explicit emotion control, as emotional signals remain entangled with speaker identity. We discover that emotion embedding emerges as a linearly decodable direction of frozen hidden states, nearly orthogonal to the direction embedding speaker identity. This inspires a plug-and-play framework DUET for emotion control over pretrained diffusion and flow-matching based TTS models. During generation, DUET unifies dual-space control to achieve fine-grained emotion intervention in a single per-step update: hidden space steering shifts generation along the target emotion direction, while mel-space guidance refines spectral details through gradients backpropagated from a differentiable vocoder. We validate DUET on five architecturally diverse pretrained TTS backbones across three datasets, where it outperforms 10 supervised state-of-the-art emotional TTS baselines across paradigms and achieves the highest human-rated emotion appropriateness. To further showcase its qualitative behavior, we deploy DUET on an Ameca humanoid robot, where it produces richly expressive emotional speech on the humanoid, demonstrating the strong potential for plug-and-play affective interaction for embodied agents. The demonstration is available at: https://anonymous.4open.science/w/duet-demo-fresh-D8E0

The rapid evolution of Embodied AI has enabled Vision-Language-Action (VLA) models to excel in multimodal perception and task execution. However, applying Reinforcement Learning (RL) to these massive models in large-scale distributed environments faces severe systemic bottlenecks, primarily due to the resource conflict between high-fidelity physical simulation and the intensive VRAM/bandwidth demands of deep learning. This conflict often leaves overall throughput constrained by execution-phase inefficiencies. To address these challenges, we propose D-VLA, a high-concurrency, low-latency distributed RL framework for large-scale embodied foundation models. D-VLA introduces "Plane Decoupling," physically isolating high-frequency training data from low-frequency weight control to eliminate interference between simulation and optimization. We further design a four-thread asynchronous "Swimlane" pipeline, enabling full parallel overlap of sampling, inference, gradient computation, and parameter distribution. Additionally, a dual-pool VRAM management model and topology-aware replication resolve memory fragmentation and optimize communication efficiency. Experiments on benchmarks like LIBERO show that D-VLA significantly outperforms mainstream RL frameworks in throughput and sampling efficiency for billion-parameter VLA models. In trillion-parameter scalability tests, our framework maintains exceptional stability and linear speedup, providing a robust system for high-performance general-purpose VLA Models.

While Reinforcement Learning (RL) enhances Large Language Model reasoning, on-policy algorithms like GRPO are sample-inefficient as they discard past rollouts. Existing experience replay methods address this by reusing accurate samples for direct policy updates, but this often incurs high computational costs and causes mode collapse via overfitting. We argue that historical data should prioritize sustaining local policy plasticity rather than simply reinforcing accuracy. We define local policy plasticity as a persistent ability to remain backtrackable and switchable with respect to recent successful trajectories: throughout training, the policy maintains non-negligible support on recently reachable alternatives, so it can smoothly reallocate probability mass when needed without being locked into a single dominant path. To this end, we propose Dynamic Jensen-Shannon Replay (DyJR), a simple yet effective regularization framework using a dynamic reference distribution from recent trajectories. DyJR introduces two innovations: (1) A Time-Sensitive Dynamic Buffer that uses FIFO and adaptive sizing to retain only temporally proximal samples, synchronizing with model evolution; and (2) Jensen-Shannon Divergence Regularization, which replaces direct replay updates with a distributional constraint to prevent premature over-commitment to a single Rank-1 trajectory. Experiments on mathematical reasoning and Text-to-SQL benchmarks demonstrate that DyJR significantly outperforms GRPO as well as baselines such as RLEP and Ex-GRPO, while maintaining training efficiency comparable to the original GRPO. Furthermore, from the perspective of Rank-$k$ token probability evolution, we show that DyJR improves policy plasticity by reducing over-reliance on Rank-1 tokens and preserving recently reachable alternatives, elucidating how specific sub-modules of DyJR influence the training dynamics.

Multi-path speculative decoding accelerates lossless sampling from a target model by using a cheaper draft model to generate a draft tree of tokens, and then applies a verification algorithm that accepts a subset of these. While prior work has proposed various verification algorithms for i.i.d rollouts, their relative performance under matched settings remains unclear. In this work, we firstly present a systematic evaluation of verification strategies across model families, tasks, and sampling regimes, and find that Traversal Verification dominates consistently, with OT-based methods lagging far behind. Our analysis uncovers that this occurs because OT-based methods achieve high multi-token acceptance near the root of the draft tree, while multi-token gains are most impactful deeper in the draft tree, where draft and target distributions diverge. Based on this insight, we propose delayed tree expansion} which drafts a partial single path, delaying the i.i.d. branching point. We show that delayed tree expansion preserves the target distribution and improves on root-node i.i.d rollouts. Further, we develop a dynamic neural selector that estimates the expected block efficiency of OT-based verification methods from draft and target features, enabling context-dependent expansion decisions. Our neural selector allows OT-based methods like SpecInfer to outperform Traversal Verification for the first time, with 5\% higher average throughput across a wide range of models, datasets, and sampling settings. Finally, we extend our approach to Traversal Verification to improve its average throughput by 26\%.

Recent DETR-based tiny object detectors adopt dynamic-query mechanisms to handle density imbalance. However, existing designs entangle two effects: how many and which queries are decoded. This entanglement obscures what actually drives the gains, thereby hindering principled dynamic-query design; additional learned predictors or heuristic filtering steps also introduce extra overhead and inevitable decision errors, which can undermine practicality and final gains. In this paper, we disentangle these effects and show that dynamic query primarily acts as budgeting---allocating decoder query capacity according to object density. Motivated by this view, we formulate a monotonic and conservative budgeting principle and propose a budgeting-based DETR (BUTR), a streamlined framework that implements dynamic query by using a lightweight, training-free budgeter to estimate an input-dependent query budget $K(x)$ from encoder confidences. BUTR applies standard Top-$K(x)$ query initialization, removing heuristic filtering, learned components, and complex selection logic. On AI-TOD-v2 and VisDrone, BUTR shows clearer density-aligned query allocation behavior than prior dynamic-query detectors and improves AP by more than 0.5, while using about half as many queries with fewer parameters and GFLOPs, requiring over 2$\times$ less training time, and delivering 4$\times$ faster inference. Additional COCO results further indicate its applicability across a broad spectrum of detection scenarios.

Leveraging the natural spatiotemporal energy decay in video diffusion offers a path to efficiency, yet relying solely on rigid static masks risks losing critical long-range information in complex dynamics. To address this issue, we propose DynamicRad, a sparse-attention framework that constrains adaptive selection using a radial locality prior. DynamicRad introduces a dual-mode strategy: static-ratio for speed-optimized execution and dynamic-threshold for quality-first filtering. To avoid online search over sparse indices, we integrate an offline Bayesian Optimization (BO) pipeline with a semantic motion router. The router maps prompt embeddings to BO-selected sparsity regimes with a single projection module. Unlike online profiling methods, our offline BO optimizes attention reconstruction error (MSE) on a proxy task and reuses the selected configurations during inference. Experiments on HunyuanVideo and Wan2.1-14B demonstrate that DynamicRad achieves a strong efficiency-quality trade-off among FlashAttention-compatible sparse-attention baselines, obtaining 1.7x-2.5x inference speedups with over 80% effective sparsity. In some long-sequence settings, the dynamic mode matches or improves over the dense baseline on automatic quality metrics, while optional Mask-Aware LoRA further improves long-horizon coherence. Code is available at https://anonymous.4open.science/r/DynamicRad-5D55/.

Tracking plasma profiles, such as electron temperature and rotation, remains a challenging task in tokamak nuclear fusion. While RL-based approaches offer a promising path for multi-input-multi-output control systems, existing frameworks overlook two key obstacles: partial observability of plasma dynamics and the diagnostic mismatch between offline training and online execution. We propose a novel offline RL framework to address these challenges. Specifically, we train the control policy to condition on the latent context that is extracted from a trained plasma dynamics model. This dynamics-informed latent context encodes useful information about operating regimes, enabling the policy to adapt as the plasma evolves. To mitigate the gap between offline and real-time observation, we learn a converter that maps plasma diagnostics reconstructed offline to their real-time-observable counterparts, thereby exposing policy training to the real-time plasma diagnostics available in live experiments. In offline evaluations, the proposed method substantially improves tracking performance and robustness across a diverse set of discharges. We report real-time deployment results on DIII-D Tokamak Fusion Facility, showing feasibility under challenging actuator conditions.

Knowledge editing aims to update outdated or incorrect knowledge in large language models without full retraining. Existing parameter-editing methods remain limited for long-sequence knowledge: single-point editors provide insufficient coverage for extended targets, while fixed chunk-based editors distribute editing effort uniformly across rigid predefined windows, failing to prioritize difficult, information-dense tokens. In this work, we show that token entropy is closely associated with both editing difficulty and semantic importance: high-entropy regions are harder to edit and often contain key factual content. Motivated by this observation, we propose DynEdit, a dynamic framework for long-sequence knowledge editing. DynEdit uses token-level entropy to guide non-uniform sliding, increasing the coverage of difficult high-entropy regions, and further applies entropy-guided token-level optimization to strengthen updates on information-dense tokens within each editing window. Experiments on three benchmarks and two backbone models show that DynEdit consistently outperforms competitive baselines across diverse long-form editing settings, achieving gains of up to +14.0 BLEU points on challenging paraphrase queries and near-saturated semantic scores on several diverse knowledge domains.

Generalist robot policies must ground language in visual observations, anticipate how actions change the scene, and reason about task goals. Existing VLA approaches typically address these capabilities separately: VLM-based policies emphasize semantic grounding, video or world models emphasize visual dynamics, and goal-conditioned methods emphasize future-state reasoning. This separation can make policies brittle under instruction shift and environment or dynamics perturbations. We introduce Dynin-Robotics, an omnimodal masked-diffusion VLA model that represents robot trajectories as partially observed multimodal token sequences. A single denoising backbone is trained to unify action generation, future-state and world modeling, trajectory-to-language goal understanding, and instruction-conditioned goal-state prediction. At inference time, the same model performs policy generation through block-wise action denoising, with optional goal-state prediction, and world-action joint decoding. We first present diagnostic analyses showing that VLM-based and video/world-model-based policies exhibit complementary failure modes, motivating a unified formulation. Our experiments show that Dynin-Robotics achieves competitive standard-task performance on LIBERO, while improving robustness on instruction-shift and perturbation settings in LIBERO-Plus and VLABench.

While novel view synthesis (NVS) for dynamic scenes has seen significant progress, reconstructing temporally consistent geometric surfaces remains a challenge. Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) offer powerful dynamic scene rendering capabilities; however, relying solely on photometric optimization often leads to geometric ambiguities. This results in discontinuous surfaces, severe artifacts, and broken surfaces over time. To address these limitations, we present DySurface, a novel framework that bridges the effectiveness of explicit Gaussians with the geometric fidelity of implicit Signed Distance Functions (SDFs) in dynamic scenes. Our approach tackles the structural discrepancy between the forward deformation of 3DGS ($canonical \rightarrow dynamic$) and the backward deformation required for volumetric SDF rendering ($dynamic \rightarrow canonical$). Specifically, we propose the VoxGS-DSDF branch that leverages deformed Gaussians to construct a dynamic sparse voxel grid, providing explicit geometric guidance to the implicit SDF field. This explicit anchoring effectively regularizes the volumetric rendering process, significantly improving surface reconstruction quality, containing watertight boundaries and detailed representations. Quantitative and qualitative experiments demonstrate that DySurface significantly outperforms state-of-the-art baselines in geometric accuracy metrics while maintaining competitive rendering performance. Codes will be publicly available.

Instruction-based image editing (IIE) models have recently demonstrated strong capability in modifying specific image regions according to natural language instructions, which implicitly requires identifying where an edit should be applied. This indicates that such models inherently perform language-conditioned visual semantic grounding. In this work, we investigate whether this implicit grounding can be leveraged for zero-shot referring image segmentation (RIS), a task that requires pixel-level localization of objects described by natural language expressions. Through systematic analysis, we reveal that strong foreground-background separability emerges in the internal representations of these models at the earliest denoising timestep, well before any visible image transformation occurs. Building on this insight, we propose a training-free framework that repurposes pretrained image editing models for RIS by exploiting their intermediate representations. Our approach decomposes localization into two complementary components: attention-based spatial priors that estimate where to focus, and feature-based semantic discrimination that determines what to segment. By leveraging feature-space separability, the framework produces accurate segmentation masks using only a single denoising step, without requiring full image synthesis. Extensive experiments on RefCOCO, RefCOCO+, and RefCOCOg demonstrate that our method achieves superior performance over existing zero-shot baselines.

While Multimodal Large Language Models (MLLMs) show promising performance in automated electrocardiogram interpretation, it remains unclear whether they perform actual step-by-step reasoning or just rely on superficial visual cues. To investigate this, we introduce ECG-Reasoning-Benchmark, a novel multi-turn evaluation framework comprising over 6,400 samples to systematically assess step-by-step reasoning across 17 core ECG diagnoses. Our comprehensive evaluation of state-of-the-art models reveals a systematic shortfall in executing multi-step logical deduction. Although models possess the medical knowledge to retrieve clinical criteria for a diagnosis, they exhibit consistently low success rates in maintaining a complete reasoning chain, primarily failing to ground the corresponding ECG findings to the actual visual evidence in the ECG signal. These results demonstrate that current MLLMs bypass actual visual interpretation, highlighting a limitation in their visual grounding capabilities and motivating robust, reasoning-centric medical AI. The code and data are available at https://anonymous.4open.science/r/ecg-reasoning-benchmark-anonymized-BB3C.

Recently, Generative Recommenders (GRs), characterized by a unified end-to-end framework, have exhibited astonishing potential in transforming the recommendation paradigm. Despite their effectiveness, we recognize that GRs are still susceptible to the long-standing issue of popularity bias that has pervaded the recommendation community. Although a few studies have attempted to extend traditional debiasing methods to GRs, their effectiveness is marginal, and the fundamental reason why GRs suffer from popularity bias remains under-explored. To bridge this gap, this study focuses on two core aspects in GRs: the optimization of generative framework and the item tokenization based on semantic index. Based on theoretical analyses, we identify that the severe popularity bias emerges from the confluence of a token-level optimization flaw and the undifferentiated property of item tokenization. Accordingly, this study develops a novel generative recommender system, called Ghost, by designing the asymmetric unlikelihood optimization and the skeleton-founded tokenization. Extensive empirical evaluations across three datasets, alongside multiple SOTA baselines, reveal that Ghost substantially alleviates popularity bias and promotes fairer recommendations, while incurring slight degradation to the overall recommendation utility.

Text-guided image editing has advanced rapidly with diffusion models and unified multimodal foundation models. However, most existing methods remain confined to single-turn settings, overlooking the more realistic scenario of multi-turn in-context editing, where users iteratively refine an image through a sequence of instructions. In this setting, a model must follow each new instruction while preserving accumulated session-level constraints, challenged by two coupled failure modes: long-context dilution, where sparse textual constraints become difficult to recover from growing interleaved image-text histories, and state contamination, where earlier editing mistakes degrade subsequent generations. We introduce Edit-R2, a novel reinforcement learning post-training framework for unified multimodal models. Edit-R2 reconstructs the operative session intent, which effectively consolidates scattered historical constraints into an explicit reasoning trace before each editing turn. It further enables multi-turn RL over both reasoning and generation through a unified objective that jointly optimizes intent reconstruction generation in discrete text space and flow-matching image generation in continuous latent space, while a trajectory filtering mechanism suppresses corrupted rollouts to stabilize training under state contamination. To support systematic evaluation, we introduce MICE-Bench, a large-scale benchmark for multi-turn in-context editing with automated metrics for instruction following (IF), content consistency (CC), and global awareness (GA) over accumulated session constraints. Experiments show that Edit-R2 substantially improves multi-turn in-context editing and achieves competitive performance compared against strong baselines.

The distributed setting for Saddle Problems (SPs) has recently emerged as a framework for modern applications in machine learning and multiagent systems. Despite its relevance, the theoretical foundations of this setting have not yet been thoroughly established. In this paper, we advance this research direction by formalizing the distributed setup for SPs and providing rigorous definitions of communication and oracle costs. Further, we prove lower bounds for any distributed gradient-span algorithm, which reveals the gap from existing methods and this theoretical limit. To this end, we provide a Decoupled Method built upon a novel multi-stage reduction that reduces the SP into a sequence of decoupled minimization tasks of residual norms. Our algorithm matches the communication lower bound, thus setting the communication complexity within the gradient-span algorithms. Moreover, it yields the first strict improvement over the long-standing oracle cost of the Extragradient method for general SPs. Finally, we study the extension of distributed SP into Variational Inequality Problem (VIP), which generalizes two-player zero-sum games to multiplayer general-sum games. We show that our Decoupled Method achieves a new state-of-the-art communication complexity for this broader class.

Pre-trained diffusion and flow policies have emerged as powerful backbones for robotics, yet adapting them to novel tasks via reinforcement learning (RL) remains computationally unstable, sample inefficient and lacks theoretical guarantees. Prior approaches that fine-tune the latent noise space bypass costly backpropagation through time but struggle with "noise aliasing," where determining the optimal noise input to the pre-trained generative model requires repeatedly refitting separate Q-value networks. This paper resolves these challenges by introducing a novel theoretical framework demonstrating that imposing a linear structure on the environment induces a Latent Noise Linear Markov Decision Process (LNL-MDP). By treating the pre-trained model as a virtual dataset, our LNL-MDP fine tune framework falls under the FineTuneRL theory with near-optimal guarantee. To translate this idealized theory into a practical methodology, we derive the Bridge Equation which eliminates the noise aliasing problem. Leveraging this insight, we propose Latent Representation Bridging for Diffusion (LARBRID), an algorithm that efficiently learns the representation space from the LNL-MDP and optimal noises to steer the pre-trained model to act optimally in given tasks. We demonstrate that LARBRID is exceptionally sample efficient, yielding up to $2\text{-}3\times$ performance increases on challenging control problems across a diverse suite of robotic locomotion and manipulation tasks.

Reliable AI evaluation is a prerequisite for measuring progress and for deployment in safety-critical settings. Unfortunately, even gathering frontier system outputs on a full suite of relevant benchmarks can be expensive, let alone adding expert review or running evaluation inside of an optimization loop. Existing approaches to sample-efficient evaluation address this concern only partially: they typically assume abundant calibration data, pretrained error-relevant embeddings, or a fixed benchmark. Meanwhile, optimization loops require repeated evaluation and the ability to surface a long tail of errors; in fact, in real world settings, the long tail of errors often drives human-in-the-loop iterative system improvement. Working in the regime in which the only available information is system inputs and one or more cheap surrogate measures of system error, we derive an optimal importance sampler for error estimation and analyze its theoretical properties. Going further, we show how the same sampler can naturally discover and rank error patterns. On standard benchmarks for frontier model question answering (MMLU-Pro) and agentic coding (TerminalBench 2.0) we show our method consistently outperforms standard Monte Carlo on estimating aggregate error, recovering failure modes and calculating their prevalence, and detecting poison agent/environment injections.

Structured sparsity extends classical sparse learning by encouraging sparsity at the level of predefined groups rather than individual parameters. This paradigm is widely adopted in modern machine learning and is typically enforced by sparsity-inducing norms. In real pipelines, the optimal group partition is often unknown a priori and does not remain static after initial training. Parameter groupings are often modified as domain knowledge or upstream representation evolves. Standard approaches require retraining the model from scratch under the new target partition, which discards the optimization effort already invested and becomes computationally prohibitive at scale. In this work, we focus on the group-wise $\ell_{p,1}$-norm, a flexible regularization family that encompasses classical group Lasso and other sparse variants. We develop a principled algorithmic framework that leverages the pre-trained optimum to fine-tune model weights toward a new desired group structure without full retraining. Technically, we decompose arbitrary group-wise transitions into a sequence of tractable primitives and derive the closed-form learning dynamics for each sub-action. Extensive experiments on real-world datasets demonstrate that our method achieves significant acceleration while provably recovering the correct optimal solution.

Asynchronous reinforcement learning (RL) is an effective paradigm for improving the training efficiency of large language model (LLM) post-training by decoupling rollout generation from policy optimization. However, this decoupling introduces stale off-policy samples generated by earlier behavior policies, leading to distribution shift and unstable policy updates. Recent replay selection methods such as D-ARL mitigate this issue by selecting variance-aware samples, but require expensive current-policy evaluation over replay responses and mainly rely on variance reduction or current-policy matching as the selection criterion. To address these limitations, we propose GA-ARL, an efficient Gradient-Aware Asynchronous Reinforcement Learning framework for LLM post-training. GA-ARL derives an analytic target distribution from the KL-regularized RL objective and introduces an advantage-weighted variant to account for both policy preference and gradient strength. During training, GA-ARL maintains a replay buffer containing samples from recent behavior policies and selects low-discrepancy samples according to the advantage-weighted analytic target, avoiding additional current-policy evaluation during replay scoring. GA-ARL outperforms SOTA asynchronous methods across mathematical, logical, and code reasoning benchmarks, achieving the best average accuracy on both Qwen3-1.7B and Qwen3-4B, with improvements of up to 6.1%. Meanwhile, compared with the SOTA D-ARL, GA-ARL reduces wall-clock training time by 25.3% on average by avoiding additional current-policy evaluation during replay selection.

The Segment Anything Model 3 (SAM 3) has achieved significant progress in *Promptable Concept Segmentation (PCS)* by processing short noun phrases to generate segmentation masks with unique instance identifiers. While SAM3 excels at single-concept segmentation, it falls short in complex, real-world environments where multiple concepts must be segmented simultaneously. In particular, SAM3 typically fails to simultaneously process concepts with semantic overlap, leading to semantic misclassifications and redundant mask predictions. Moreover, as the number of concepts increases, the independent inference paradigm introduces prohibitive inference latency. To address these limitations, we propose the **D**ense **C**ompetitive **R**epresentations **SAM** (**DCR-SAM**) framework. By decoupling cross-modal interactions, the proposed method alleviates the inference latency. To resolve semantic misclassifications, DCR-SAM dynamically aligns textual concepts with visual representations and incorporates a dense competition head to enforce explicit inter-class competition. Furthermore, a learnable background token is applied to absorb non-target objects dynamically. These mechanisms effectively suppress overlapping predictions and generate high-fidelity masks. Extensive experimental evaluations across multiple benchmarks demonstrate that the proposed framework: I) achieves a 2.3\% $\sim$ 10.7\% performance improvement over state-of-the-art methods, II) delivers a 33$\times$ inference speedup compared to the native SAM 3, and III) exhibits strong robustness across diverse scenarios. All code will be released.

Scaling long-context capabilities is crucial for Large Language Models (LLMs). However, real-world data contain a large number of sequences with heterogeneous lengths. Existing training libraries for LLMs rely on static parallelism strategies, which suffer from severe load imbalance, redundant communication, and suboptimal hardware utilization under data heterogeneity. In this work, we propose Flexible Context Parallelism (FCP), an efficient parallelism strategy that adaptively reconfigures communication groups and context parallelism degrees during LLM training. We generalize more flexible non-power-of-two parallelism degrees and develop a polynomial-time algorithm to generate near-optimal parallelism strategies with only millisecond-level overhead per training batch. FCP is able to maintain high hardware efficiency even under extreme data heterogeneity. Experimental results demonstrate that FCP significantly outperforms Megatron-LM and DeepSpeed in both LLM and MLLM training, achieving up to 1.47 $\times$ speedup in average throughput while maintaining near-linear scaling efficiency across large-scale clusters. For extremely unbalanced batches, FCP even achieves 2.24 $\times$ speedup.

Streaming audio-visual target speaker extraction (AVTSE) is essential for latency-sensitive applications. However, current causal systems face two coupled gaps. First, standard benchmarks fix the number of speakers at two and ignore the near-/far-field interference, reverberation, and sudden noise of real scenes. Second, high-performance causal architectures are too heavy for edge deployment. Lightweight alternatives try to reclaim capacity by repeatedly invoking a small separator, which offsets the savings. To address the first gap, we release RealSSA, a realistic AVTSE benchmark consisting of RealSSA-Sim and RealSSA-Real. RealSSA-Sim provides controllable simulated scenes with 2-6 dynamic near-/far-field speakers across four scene types, while RealSSA-Real provides a held-out real-recorded evaluation set. To address the second gap, we propose Falcon, a causal AVTSE method. Its Separator performs multi-scale time-frequency separation in a single encoder--decoder pass. We train Falcon with an encoder-space speaker contrastive loss. This loss suppresses near-field leakage at zero inference cost and transfers across backbones. Falcon achieves state-of-the-art extraction quality on RealSSA-Sim, LRS2-2Mix, LRS3-2Mix, and VoxCeleb2-2Mix. Compared with the causal AV-TFGridNet baseline, it cuts parameters by 91.9\%, computation by 20.2$\times$, and GPU inference latency by 11.1$\times$. Code and demo are available at https://submission-falcon.github.io/demo/.

Vision-Language-Action (VLA) models have emerged as a powerful paradigm for generalist robotic control. However, their high computational cost and limited control frequency hinder real-time robotic manipulation, especially when large vision-language backbones and iterative action heads are executed at every control step. Existing VLA acceleration methods often optimize individual components or rely on fixed acceleration rules, treating different control steps with largely fixed computation and overlooking the non-uniform reasoning demands of sequential embodied control. Inspired by human motor control, where cognitive and feedback resources concentrate on goal-sensitive stages, we argue that VLA models should learn when to invest full computation and when to reuse prior computation. To this end, we propose ElegantVLA, a plug-in phase-adaptive inference framework that accelerates VLA models through intra-model dynamic compute scheduling. ElegantVLA introduces a lightweight scheduler that observes temporal representation similarity, robot-motion cues, and episode progress to jointly allocate computation across the vision encoder, LLM, and action head. For perception-language reasoning, the scheduler selects a five-level Vision--LLM compute mode, from full recomputation to multi-step temporal reuse, based on visual-language representation stability. For action generation, it selects a three-level denoising mode, reusing intermediate denoising states during stable motion while preserving full refinement for goal-sensitive stages. By coordinating these decisions, ElegantVLA provides a general acceleration framework for modern VLA pipelines with explicit action-generation modules, without modifying or retraining the base model. Extensive experiments on GR00T, CogACT, and real-world tasks show that ElegantVLA preserves or improves task success while substantially accelerating inference. On GR00T, it achieves up to 2.55$\times$ average speedup. On CogACT, it delivers a 3.77$\times$ average speedup. In GR00T-based real-world experiments across six tasks, it reduces computation by 2.18$\times$ and increases control frequency from 13.8 Hz to 26.3 Hz.

Biological neural networks (BNNs) have been established as a powerful and adaptive substrate that offer the potential for incredibly energy and data efficient information processing with distinct learning mechanisms. Yet a core challenge to utilizing BNN for neurocomputation is determining the optimal encoding and decoding mechanisms between the traditional silicon computing interface and the living biology. Here, we propose an Embodied Neurocomputation framework as a systems-level approach to this multi-variable optimization encoding/decoding problem. We operationalize this approach through the first large-scale parameter optimization of encoding configurations for a BNN agent performing closed-loop navigation along an odor-style gradient in a simulated grid-world. Despite the relative simplicity of the task, the biological interactions gave rise to a massive multi-combinatorial search space for optimal parameters. By considering how the components of the system are interconnected and parameterized, we evaluated approximately 1,300 parameter combinations, over 4,000 hours of real-time agent-environment interactions, to identify 12 configurations that consistently demonstrated learning across multiple episodes. These configurations achieved significantly higher task performances than optimized silicon-based DQN agents under the same interaction budget. These findings represent an initial step toward robust and scalable goal-oriented learning using BNNs. Our framework establishes a foundation for applying task-driven neurocomputing and supports the development of field-wide benchmarks. In the long term, this work supports the development of hybrid bio-silicon architectures capable of efficient, adaptive and real-time computation, including the potential for robotic control applications.

Recent advances in 3D object generation have enabled the creation of high-fidelity 3D assets from 2D images. This 2D-to-3D capability offers a promising path for alleviating 3D data scarcity, which remains a key bottleneck for scalable simulation in robotics and AR/VR applications. However, existing 3D generation methods primarily emphasize visual plausibility and geometric fidelity, with limited consideration of object-level interaction and structural understanding that are critical for downstream embodied AI tasks. To bridge this gap, we introduce EmbodiedObject, a unified framework for embodied 3D object understanding built upon latent representations produced by modern 2D-to-3D generative models such as SAM3D. We first systematically analyze the part-level locality of these latent representations against existing unified 3D point encoders and show that they naturally preserve rich structural semantics suitable for object understanding. Motivated by this observation, we then design a unified DiT-style decoder that operates directly on the latent representation and supports multiple object-level tasks, including open-vocabulary 3D affordance prediction, 3D part segmentation, and 3D articulation estimation. Given a real-world 2D image containing an object of interest, EmbodiedObject extends 2D-to-3D generation beyond visual reconstruction by jointly reasoning about affordance, articulation, and part semantics. This transforms generated 3D assets into actionable object representations suitable for downstream embodied AI applications. Our source-code will be open-sourced.

Knowledge Graphs (KGs) are structured knowledge repositories containing entities and relations between them. In this paper, we study the problem of automatically updating KGs over time in response to evolving knowledge in unstructured textual sources. Addressing this problem requires identifying a wide range of update operations based on the state of an existing KG at a given time and the information extracted from text. This contrasts with traditional information extraction pipelines, which extract knowledge from text independently of the current state of a KG. To address this challenge, we propose a generic and extensible pipeline that pairs textual passages with the KG edit operations they induce on a given KG snapshot. We instantiate this pipeline on the Wikidata knowledge graph and the English Wikipedia corpus to create EMERGE, a dataset of 233K Wikipedia passages associated with a total of 1.19 million KG edits across seven yearly Wikidata snapshots from 2019 to 2025. Our experimental results highlight key challenges in updating KG snapshots based on emerging textual knowledge, particularly in integrating knowledge expressed in text with the existing KG structure. These findings position the dataset as a valuable benchmark for future research. The code and dataset are available at https://github.com/klimzaporojets/emerge.

Test-time sampling can improve Vision-Language-Action (VLA) policies, but only if the model can efficiently select a good action from multiple candidates. Prior work often relies on separately trained external verifiers, adding extra models and inference overhead. We propose \textbf{Token Probability Bucketing (TokenPB)}, a self-verification approach that discretizes continuous action chunks into bucket tokens and co-trains a single VLA checkpoint with a joint objective: a flow-matching loss for action generation and an auxiliary bucket loss $\mathcal{L}_{\text{bucket}}$ that teaches the VLM backbone to model observation-conditioned distributions over bucket-token sequences. At inference, TokenPB samples $M$ candidates, scores each via teacher-forced evaluation under the learned distribution, and selects the best---enabling efficient batched scoring with observation-prefix KV-cache reuse. We provide theory connecting the training objective to a monotonic ranking property and bounds on expected selection regret that separate candidate-set effects from calibration. Across multiple backbones and benchmarks, TokenPB yields consistent success-rate improvements over single-sample inference from the same checkpoint (+2.7 in simulation, +17.1 in real-world, +16.4 under out-of-distribution shifts), and further improves over external-verifier baselines under the same sampling budget.

Denoising and score estimation are classically linked through Tweedie’s formula, which relates the posterior mean under Gaussian corruption to the Stein score of the noisy marginal. In this work, we extend this perspective beyond Gaussian noise to a broad class of elliptical, energy-based noise distributions, with particular emphasis on generalized Gaussian corruptions. We derive the Energy–Tweedie identity: when the denoising posterior is viewed through the lens of proper scoring rules, the path derivative of a matched, possibly non-Euclidean energy score recovers the Stein score of the noisy marginal. Thus, the familiar correspondence between Gaussian noise, posterior means, squared loss, and Tweedie’s formula is lifted to a distributional correspondence between generalized Gaussian noise, full posterior laws, Mahalanobis energy scores, and the Energy–Tweedie identity. Among its consequences, this identity gives a posterior-sample-based route to score estimation, yields a principled criterion for noise-parameter calibration, and supplies a score-based perspective on recent diffusion-style generative methods trained with scoring rules.

While agentic frameworks have advanced repository-level code localization, their reliance on static snapshots often leads to a $\textit{traceability gap}$—the loss of implicit links between issue symptoms and code implementation. To bridge this information deficit, we introduce $\textbf{T}$raceability $\textbf{A}$ugmented $\textbf{Co}$de Localization (TACO), a framework that recovers retrievable and usable traceability from repository evolution to guide localization agents. TACO offline crystallizes historical pull requests into a dual-index knowledge base through a prompt auto-tuning technique, capturing high-value semantic hints and architectural rationales. In the online phase, TACO employs a synergistic dual-track retrieval workflow, cross-validated by an arbiter and aligned temporally to handle version drift. Extensive evaluations on SWE-bench Lite and SWE-bench Verified across five state-of-the-art agentic baselines demonstrate that TACO significantly improves exact-match localization accuracy (Acc@1) by an average of 11.29\%, 13.20\%, and 12.87\% at the file, module, and function levels, respectively. Moreover, TACO significantly streamlines the exploration process, achieving an average reduction of 29.2\% in token usage and 30.4\% in monetary costs, with minimal offline maintenance overhead.

LLM agents are increasingly expected to carry out end-to-end workflows, producing complete artifacts from high-level user instructions. To meet enterprise needs, frontier AI labs have developed agents that can construct entire spreadsheets from scratch. This capability is especially relevant in finance, where core workflows such as financial modeling, forecasting, and scenario analysis are commonly conducted through spreadsheets. Yet, existing spreadsheet benchmarks do not measure this advanced capability, focusing instead on question-answering or single-formula edits. To address this gap, we provide one of the first evaluations of agents on end-to-end spreadsheet tasks, focusing on economically critical financial workflows such as modeling and scenario analysis. Since deliverables therein are routinely reviewed and revised by multiple stakeholders, judging their quality necessarily involves high-level criteria such as readability or ease of modification. To reflect the multidimensional nature of solution quality, we develop an evaluation taxonomy comprising three dimensions: Accuracy, Formula, and Format, each comprising fine-grained criteria that reflect professional standards. The Claude family leads the benchmark and produces the most professional-looking outputs in our qualitative review, but even the strongest agents frequently fall short of professional finance standards and degrade sharply as the difficulty increases beyond a few chained calculations. This suggests that current agents are not yet able to reliably produce professional-quality spreadsheets at the level of complexity real-world workflows demand.

Recent text-to-image (T2I) generators can synthesize realistic images, but still struggle with compositional prompts involving multiple objects, counts, attributes, and relations. We introduce EPIC (Efficient Predicate-Guided Inference-Time Control), a training-free inference-time refinement framework for compositional T2I generation. EPIC casts refinement as predicate-guided search: it parses the original prompt once into a fixed visual program of object variables and typed predicates, covering checkable conditions such as object presence, counts, attributes, and relations. Each generated or edited image is verified against this program using visual evidence extracted from that image. An image is judged to satisfy the prompt only when all predicates are satisfied; otherwise, failed predicates decide the next step, routing local failures to targeted editing and global failures to resampling while the fixed visual program remains unchanged. On GenEval2, EPIC improves prompt-level accuracy from 34.16% for single-pass generation with the base generator to 71.46%. Under the same generator/editor setting and maximum image-model execution budget, EPIC outperforms the strongest prior refinement baseline by 19.23 points while reducing realized cost by 31% in image-model executions, 72% in MLLM calls, and 81% in MLLM tokens per prompt.

Mapping fine-scale 3D chromatin landscapes is critical for understanding gene regulation, yet the resolution of Hi-C data remains fundamentally constrained by both physical bottlenecks and sequencing costs. Although deep learning has shown promise in super-resolving Hi-C data, existing methods suffer from absolute coordinate loss during localized patching and a lack of structural and epigenomic priors in unconditioned generation. We propose a novel epigenomics-guided flow matching framework EpiFlow for 3D genome super-resolution. EpiFlow introduces two key innovations: (1) an EpiCond module that deterministically maps 1D epigenomic features to 2D spatial constraints via orthogonal expansion with built-in positional grounding, avoiding quadratic attention complexity; and (2) a Toeplitz-regularized classifier-free guidance mechanism that enforces distance-decay priors using a learnable symmetric prototype matrix. Extensive evaluation shows that our method achieves state-of-the-art cross-cell-type loop detection, improving recall by $2.8\times$ over raw Hi-C on unseen H1-hESC (Loop $F_1$ of $0.325$ vs. $0.199$) and outperforming all baselines on same-cell HFFc6 ($F_1=0.520$). Aggregate Peak Analysis (APA) confirms biological validity, and strong generalization to a held-out cell line demonstrates that the model learns transferable principles of 3D genome folding rather than cell-specific biases.

Referring expression segmentation (RES) aims to produce a pixel-level mask for the object described by a free-form natural language expression, in either images (RIS) or videos (RVOS). Existing zero-shot approaches reduce the cost of mask-language annotation but still rely on auxiliary training stages or multi-model pipelines. The recently released SAM 3 introduces native text-prompted segmentation within a single foundation model. However, directly applying it to RES generates unexpected weak performance. We trace this gap to an attention sink in SAM 3's multimodal fusion encoder, where the start-of-text token absorbs on average $54.8\%$ of each image patch's cross-attention mass, leaving content tokens with limited influence on the fused visual features. Suppressing the sink alone is insufficient, as the released attention mass spreads without direction; closing the gap requires both suppressing the sink and providing patch-dependent spatial guidance that routes attention to semantically relevant content tokens. Building on these findings, we introduce eSAM, a fully training-free framework that edits SAM~3's cross-attention without any parameter updates. With Attention Mask Editing (AME), eSAM edits the attention map, suppressing the sink while injecting a CLIPSeg-derived spatial prior that directs released attention toward relevant content tokens. Further with Value Tensor Editing (VTE), eSAM rescales the value tensors, amplifying the magnitude with which content tokens contribute to the fused output. On both RIS (RefCOCO/+/g) and RVOS (Ref-DAVIS17, Ref-YouTube-VOS, MeViS) benchmarks, eSAM achieves new state-of-the-art results among training-free methods.

Biological multimodal large language models (MLLMs) have emerged as powerful foundation models for scientific discovery. However, existing models are specialized to a single modality, limiting their ability to solve inherently cross-modal scientific problems. While model merging is an efficient method to combine the different modalities into a unified MLLM, existing methods rely on input-agnostic parameter space heuristics that fail to faithfully capture modality specialization. To overcome this limitation, we propose the Embedding-Signal-based MLLM Merging (ES-Merging), a framework that estimates merging coefficients from embedding space signals, moving the merging paradigm from the parameter signals to the embedding signals. ES-Merging exploits coarse-grained and fine-grained signals from embedding space to estimate the layer-wise and element-wise merging coefficients, respectively, which are jointly combined for complementary coefficient estimation. Through extensive experiments, we demonstrate that ES-Merging outperforms existing merging methods not only on the cross-modal reasoning but also on the single-modal knowledge preservation, establishing that embedding space signals provide a principled and effective foundation for MLLM merging.

Membership inference attacks (MIAs) have emerged as the standard tool for evaluating the privacy risks of AI models. However, state-of-the-art attacks require training numerous, often computationally expensive, reference models, limiting their practicality. We present a novel approach for estimating model-level vulnerability to the Likelihood Ratio Attack (LiRA), the strongest available attack, directly from the train and test loss distributions of the target model and without training any reference models. We show that LiRA's per-sample signal decomposes into a variance-ratio term and a residual mean-shift term, with the relative contribution of each determined by how much training collapses model uncertainty at the trained sample. This places models on a continuum, with different regimes calling for different reference-free loss-based statistics as proxies for LiRA TPR. The shapes of the loss distributions themselves indicate which proxy applies. We instantiate the framework with two natural proxies. At the heavy-tailed end, the LOSS attack TNR predicts LiRA TPR@FPR=$10^{-3}$ with RMSE 0.03 across 9 image classification architectures and 4 datasets, outperforming low-cost reference-model attacks such as RMIA. At the symmetric end, the LOSS attack AUC predicts LiRA TPR with RMSE 0.01 across five GPT-2 sizes from 10M to 1B parameters. We also show these proxies to outperform both low-cost (few reference models) attacks such as RMIA and other measures of distribution difference.

LLM agents are increasingly expected to operate as general-purpose systems that resolve real-world user requests, yet their dynamic scaling behavior in realistic environments remains poorly understood. In this paper, we systematically investigate two principal test-time scaling axes of LLM agents: sequential scaling through extended interaction and parallel scaling through trajectory sampling. We first introduce a realistic benchmark that provides one unified framework for evaluating LLM agents across search, coding, reasoning, and tool-use domains, more faithfully reflecting the heterogeneity of real-world deployments. Evaluating ten leading LLM agents reveals substantial performance degradation when transitioning from domain-specific evaluations to this realistic setting. Building on this foundation, we progressively scale test-time compute along fine-grained increments to characterize the performance upper bound. We find that neither scaling axis can consistently yield meaningful gains from additional test-time compute in realistic environments, a phenomenon we attribute to two fundamental limitations: the scaling plateau that bottlenecks sequential scaling and the verification gap that undermines parallel scaling. Code and data will be released publicly.

Time series foundation models (TSFMs) are increasingly used for probabilistic forecasting across diverse domains, yet the reliability of their uncertainty estimates remains poorly understood. In this work, we study uncertainty calibration in probabilistic TSFMs from a broader methodological and empirical perspective. We introduce a taxonomy of how TSFMs represent predictive uncertainty and develop a fine-grained evaluation framework for calibration in multi-horizon forecasting that distinguishes pooled, horizon-wise, and stronger conditional notions of calibration. Using this framework, we conduct a comprehensive empirical evaluation of popular TSFMs across diverse datasets and show that current models exhibit distinct and systematic patterns of miscalibration that standard metrics often obscure. We further analyze common post-hoc recalibration methods and characterize which forms of calibration they can improve and which tend to persist. Our results provide practical guidance for evaluating and improving uncertainty calibration in TSFMs and highlight important limitations of current practice.

First-person dynamic spatial reasoning requires models to track continuous motion and precise geometric structure, but the quadratic attention cost of Transformer-based Video-LLMs makes dense visual tokens computationally expensive. Existing token pruning paradigms predominantly rely on discrete static snapshots, failing to preserve the physical dynamics essential for reasoning. We propose Event Cascade Pruning (ECP), to our knowledge the first training-free framework that leverages the high-frequency motion cues from event cameras as a continuous event-guided motion prior to guide token selection. ECP combines three stages: Event-Triggered Causal Sampling to anchor motion-informative keyframes, Event-guided Motion Saliency Filtering to remove static or low-motion regions, and Event-Attention Ranking Fusion to calibrate spatial attention with motion salient dynamics. With 80\% visual token reduction, ECP outperforms the full-token baseline (37.62\% vs. 36.31\%) while achieving 1.89× inference speedup and 52\% GFLOPs reduction. We further introduce ESR-Real, the first real-world RGB-event benchmark for first-person spatial reasoning, where ECP improves accuracy by 2.68\% over full-token baselines.

With LLM watermarking already being deployed commercially, practical applications increasingly require multibit watermarks that encode more complex payloads, such as user IDs or timestamps, into the generated text. In this work, we propose a fundamentally new approach for multibit watermarking: introducing binomial encoding to directly encode every bit of the payload at every token position. We complement our approach with a stateful encoder that during generation dynamically redirects encoding pressure toward underencoded bits. Our evaluation against 8 baselines on up to 64-bit payloads shows that our scheme achieves superior message accuracy and robustness, with the gap to baseline methods widening in more relevant settings (i.e., large payloads and low-distortion regimes). At the same time, we challenge prior works’ evaluation metrics, highlighting their lack of practical insights, and introduce per-bit confidence scoring as a practically relevant metric for evaluating multibit LLM watermarks.

Fine-tuning MLLMs for Video Temporal Grounding (VTG) often improves in-domain performance but degrades sharply under domain shift. In this work, we find that this failure is primarily driven not just by unseen query concepts, but by visual domain shift, which prevents the model from coupling its learned temporal localization knowledge with its inherent entity-attention capability. To address this, we introduce EVIDENT, a parameter-efficient adaptation framework that anchors temporal grounding in the inherent entity-attention of pre-trained MLLMs by routing VTG adaptation through explicit visual entity evidence. EVIDENT consists of three components: (i) an Entity Bottleneck Adapter that transforms dense visual tokens into compact entity-level slots, (ii) an Entity-Binding Distillation loss that instills objectness priors into the semantically unstructured MLLM visual space, guiding each slot to bind to a coherent entity, and (iii) an Entity-to-eVidence gating mechanism that leverages the captured entities as evidence, steering the model to localize moments containing query-relevant entities. Together, these components enable VTG fine-tuning to rely on entity-grounded evidence rather than brittle dataset shortcuts. Experiments on cross-domain VTG benchmarks show that EVIDENT consistently improves out-of-domain robustness while preserving competitive in-domain performance with modest parameter overhead. These results suggest that entity-level grounding is an effective inductive bias for generalizable temporal localization.

Referring Remote Sensing Image Segmentation (RRSIS) aims to segment target objects from remote sensing images based on natural language descriptions, requiring precise vision–language alignment under significant ambiguity. Despite the strong generalization ability of Segment Anything Model 3 (SAM3), the large spatial extent and varying scales of targets in remote sensing images intensify the ambiguity of referring expressions. This poses a major challenge for fine-grained segmentation. To address the challenge, we propose EviSAM3, an evidence-driven parameter-efficient adaptation framework that mimics human decision-making by dynamically accumulating partial evidence. Specifically, EviSAM3 progressively integrates cross-modal evidence through an evidence memory bank, where each incoming piece is aggregated via momentum-based updates. As evidence accumulates, the model performs iterative evidence rectification to resolve ambiguity under incomplete observations. Meanwhile, the contributions of different evidence are adaptively modulated based on their diagnostic relevance, allowing more informative evidence to dominate the reasoning process and guide the final prediction. Extensive experiments demonstrate that EviSAM3 consistently outperforms state-of-the-art methods, particularly in challenging scenarios with high ambiguity and significant scale variation. The code is available in the supplementary material.

Video temporal grounding (VTG) takes an untrimmed video and a natural-language query as input and localizes the temporal moment that best matches the query. Existing methods rely on large, task-specific datasets requiring costly manual annotation. We introduce EvoGround, a framework of two coupled self-evolving agents, a proposer and a solver, that learn temporal grounding from raw videos without any human-labeled data. The proposer generates query--moment pairs from raw videos, while the solver learns to ground them and feeds back signals that improve the proposer in return. Through this self-reinforcing reinforcement-learning loop, the two agents are initialized from the same backbone and mutually improve across iterations. Trained on 2.5K unlabeled videos, EvoGround matches or surpasses fully supervised models across multiple VTG benchmarks, while emerging as a state-of-the-art fine-grained video captioner without manual labels.

Diffusion and flow-based models learn powerful data priors by training a denoiser to reverse Gaussian corruption. To use this prior to solve a linear inverse problem, one needs to sample from the posterior, but the score that the prior provides is the unconditional score, not the posterior score. Existing methods either steer a fixed pretrained denoiser with approximate measurement-matching corrections, or train a conditional restoration model that abandons the denoising structure of the prior. We derive the exact posterior score in closed form for linear Gaussian inverse problems under general Gaussian interpolants, and show that posterior sampling reduces to a denoising problem at an operator-dependent shifted pivot under an anisotropic noise covariance. We turn this identity into Exact Posterior Score (EPS), a denoising training objective that preserves the input/output structure of standard pretraining and can therefore be trained from scratch or fine-tuned from a pretrained denoiser. At inference, EPS uses the same sampler as the underlying backbone, with no likelihood gradients or projections. We evaluate EPS on five linear inverse problems across FFHQ and ImageNet, where it outperforms training-free and training-based baselines on fidelity, perceptual, and distributional metrics, while using roughly an order of magnitude fewer denoiser evaluations than gradient-based posterior samplers.

Forecasting spatio-temporal human mobility is essential for urban infrastructure optimization, yet dense traffic sensing remains prohibitively expensive for many cities. Existing cross-city transfer methods and recent spatio-temporal foundation models usually rely on historical target-side traffic sequences as the prediction context, which limits their applicability to under-instrumented, graph-missing, or newly monitored cities. We study exogenous-driven cross-city transfer, where target traffic must be predicted from static and dynamic exogenous urban context without historical flow input. This exogenous-only observability removes the direct endogenous state observation used by flow-driven predictors and turns cross-city forecasting into latent traffic-generating mechanism recovery under city shift. The key challenge is to recover traffic-state information from indirect external evidence while organizing cross-city response patterns into a mechanism space that can adapt to target variation without destabilizing transferable structure. To address this challenge, we propose \textbf{ExoC2T}, a plasticity-inspired spatio-temporal learning framework. ExoC2T uses an Exogenous-Infused Spatio-Temporal Network (\textbf{EXIST}) to build an exogenous mechanism substrate, where semantic relations, temporal regimes, and environment-conditioned responses are organized into a shared prototype geometry. On this substrate, Stability-Plasticity Adaptation separates a stable exogenous-to-flow core from an adaptive city-conditioned deviation, infers a support-conditioned latent posterior from limited target evidence, and controls target risk through invariant discrepancy reduction and posterior correction. The resulting model preserves transferable exogenous-to-flow structure while regulating target-specific modulation through posterior-dependent plasticity. Extensive experiments on a multi-city benchmark show consistent improvements over strong forecasting and transfer baselines, demonstrating the value of exogenous-driven transfer for low-cost and scalable urban computing.

Medical agent systems are increasingly expected to support interactive clinical decision making rather than only static question answering. In such settings, effective agents must reuse prior experience across evolving cases, yet existing memory mechanisms often retain raw historical traces that are redundant, noisy, and difficult to govern. More importantly, they rarely distinguish which memories are truly useful for future reasoning. This limits their ability to accumulate compact and reliable experience for long-horizon clinical reasoning. To close this gap, we propose SkeMex, a post-deployment self-evolution framework that improves medical agents through a skill-based memory without updating model weights. SkeMex distills informative interaction trajectories into structured skills that encode reusable procedural knowledge, and organizes them into a multi-branch repository spanning general, task-specific, and action-level experience. To determine which memories should be reused and retained, SkeMex estimates context-dependent utility from environment feedback and uses it to guide value-aware retrieval and repository governance. A closed-loop ``Read--Write--Assess--Govern" lifecycle further supports continual evolution by writing new skills, updating utilities, promoting useful memories, and removing harmful entries. Experiments across diverse clinical tasks show that SkeMex consistently outperforms representative memory-based agents in both offline and online settings. It also generalizes across model backbones and supports transferable skill memory. All data and code will be released publicly.

Diversifying protein sources away from animal agriculture is critical for climate change mitigation and food security, but formulating sustainable protein sources that match or improve upon the properties of their animal-based counterparts remains an expensive trial-and-error process. We frame this challenge as high-dimensional black-box optimization with sparse approximate solutions and formalize Expert-Guided Bayesian Optimization (EGBO), in which an expert, e.g. a human or LLM, selects a low-dimensional subspace for BO and may adaptively expand it over time. We decompose EGBO's suboptimality into a selection gap and an optimization gap, and characterize the coverage–dimension tradeoff governing when expert guidance helps. To support in silico prototyping before costly real-world deployment, we introduce FormulateBench, a suite of 24 plant-based formulation tasks, on which LLM-guided EGBO outperforms all tested baselines. When deployed to optimize two plant-based dairy products, EGBO improves utility, as assessed by a trained human panel, by 29\% and 26\% in 10 iterations each. In a comparison with a professional human food scientist given the same time budget, EGBO achieved near-perfect utility of 0.992, vs. 0.850 for the food scientist.

Reinforcement learning from human feedback (RLHF) typically assumes a static or non-strategic reward model (RM). In iterative deployment, however, the policy generates the data on which the RM is retrained, creating a feedback loop. Building on the Stackelberg game formulation of this interaction, we derive an analytical decomposition of the policy's true optimization gradient into a standard policy gradient and a parameter-steering term that captures the policy's influence on the RM's future parameters. We show that standard iterative RLHF, which drops this steering term entirely, suffers from alignment collapse: the policy systematically exploits the RM's blind spots, producing low-quality, high-reward outputs whose feedback reinforces the very errors it exploits. To mitigate this, we propose foresighted policy optimization (FPO), a mechanism-design intervention that restores the missing steering term by regularizing the policy's parameter-steering effect on RM updates. We instantiate FPO via a scalable first-order approximation and demonstrate that it prevents alignment collapse on both controlled environments and an LLM alignment pipeline using Llama-3.2-1B.

Current large reasoning models (LRMs) have shown strong ability on challenging tasks after reinforcement learning (RL) based post-training. However, previous work mainly focuses on English reasoning in expectation of the strongest performance, despite the demonstrated potential advantage of multilingual thinking, as well as the requirement for native thinking traces by global users. In this paper, we propose ExpLang, a novel LLM post-training pipeline that enables on-policy thinking language selection to improve exploration and exploitation during RL with the use of multiple languages. The results show that our method steadily outperforms English-only training with the same training budget, while showing high thinking language compliance for both seen and unseen languages. Analysis shows that, by enabling on-policy thinking language selection as an action during RL, ExpLang effectively extends the RL exploration space with diversified language preference and improves the RL exploitation outcome. The method is orthogonal to most RL algorithms and opens up a new perspective on using multilinguality to improve LRMs.

While federated reinforcement learning has traditionally guaranteed high efficiency in homogeneous environments, collaboration is often hindered when agents possess distinct environments and goals. In this paper, we flip this paradigm, framing reward heterogeneity not as a hurdle to consensus, but as a structural asset that catalyzes exploration. We first propose Personalized Federated Upper Confidence Bound Value Iteration (PF-UCBVI), which achieves optimal linear speedup by decoupling shared dynamics from personalized objectives. To eliminate the risks of explicit exploration, we then introduce Personalized Federated Exploration-Free Value Iteration (PF-EFVI), a purely greedy algorithm that leverages reward diversity to ensure state-action coverage. We prove that PF-EFVI attains logarithmic regret without explicit exploration bonuses under sufficient reward diversity. Our results show that agent disagreement is a vital resource that shifts the exploration burden from temporal complexity to spatial diversity, enabling safe and efficient collective learning.

Multimodal Retrieval-Augmented Generation (RAG) helps mitigate hallucinations in Vision-Language Models (VLMs) by grounding generations in external knowledge bases. However, this externalized memory also introduces privacy risks, as external queries may reveal signals about sensitive visual records in the retrieval corpus, such as medical images, scanned contracts, and proprietary business documents. This exposes a retrieval-corpus privacy risk: private records may be detectable through interactions with multimodal RAG systems. To study this risk, we propose the Semantic Degradation Attack (SDA), a two-query method for exposing private corpus leakage in multimodal RAG by testing how strongly generated responses depend on retrieved evidence. SDA constructs transferable retrieval-disrupting perturbations using local surrogate visual encoders. By measuring the drop in the semantic similarity of the VLM's generated descriptions before and after perturbation, SDA can distinguish whether a target image-caption record exists in the private retrieval corpus. Extensive experiments demonstrate that SDA consistently detects private corpus leakage more effectively than existing baselines on two image-caption datasets across five popular VLMs.

Extremely sparse-view computed tomography (xSV-CT) is critical when scanning time, motion, or radiation tolerance limits each scan to only a few projections. In many such scenarios, the same acquisition limits also constrain dataset construction, making pre-reconstructed 3D volumes from dense scans unavailable. This exposes a gap in current sparse-view approaches: Optimization-based methods, which use the sparse measurements directly but lack data-driven priors to suppress artifacts, and learning-based methods, which use strong priors but usually require densely scanned training volumes acquired outside the xSV-CT regime. We study this train-sparse / infer-sparse setting by learning a pose-aware diffusion prior from one 2D projection per training sample, then guiding diffusion sampling with measured-view fidelity, shared-volume reprojection, and projection-physics consistency for tomographic reconstruction. On two public CT datasets across 1-, 4-, and 8-view settings, the proposed method achieves the best novel-view synthesis scores and best or competitive reconstruction quality compared with baselines that use dense-volume supervision.

Diffeomorphisms provide a flexible, topology-preserving mathematical tool for modeling complex spatial deformations, and are used in various scientific fields. However, simultaneously achieving high expressivity and computational efficiency remains challenging. Continuous Piecewise-Affine Based (CPAB) transformations offer an attractive solution by parameterizing a family of diffeomorphisms via continuous velocity fields that are piecewise affine *w.r.t.* a chosen tessellation of the domain. Importantly, $d=\dim(\theta)$, the dimension of the parameter vector $\theta$, depends on the fineness of the tessellation rather than on the data resolution. Despite this compact parameterization, CPAB optimization has long been hindered by the tight coupling between trajectory integration and computing gradients *w.r.t.* $\theta$. We introduce **FG-CPAB**, a factorized-gradient formulation for scalable CPAB transformations. By separating integration from parameter-space projection, our method reduces gradient computation complexity from $(\mathcal{O}(dTN))$ to $(\mathcal{O}(TN + d C))$, where $(T)$ is the number of integration steps, $(N)$ is the number of transformed points, and $(C)$ is the number of tessellation cells. This reformulation yields huge speedups (e.g., $\approx 10^4\times$ in 2D), substantial memory savings (e.g., $\approx 500\times$ reduction in peak GPU memory usage in 2D), and dramatically higher expressiveness. This makes, for the first time, fine-tessellation CPAB practical in 2D and 3D, unlocking the potential of highly-expressive parametric diffeomorphisms.

Confirmation bias, the tendency to seek evidence that supports rather than challenges one's belief, hinders one's reasoning ability. We examine whether large language models (LLMs) exhibit confirmation bias by adapting the rule-discovery study from human psychology: given a sequence of three numbers (a "triple"), an agent engages in an interactive feedback loop where it (1) proposes a new triple, (2) receives feedback on whether it satisfies the hidden rule, and (3) guesses the rule. Across eleven LLMs of multiple families and scales, we find that LLMs exhibit confirmation bias, often proposing triples to confirm their hypothesis rather than trying to falsify it. This leads to slower and less frequent discovery of the hidden rule. We further explore intervention strategies (e.g., encouraging the agent to consider counter examples) developed for humans. We find prompting LLMs with such instruction consistently decreases confirmation bias in LLMs, improving rule discovery rates from 42\% to 56\% on average. Lastly, we mitigate confirmation bias by distilling intervention-induced behavior into LLMs, showing promising generalization to a new task, the Blicket test. Our work shows that confirmation bias is a limitation of LLMs in hypothesis exploration, and that it can be mitigated via injecting interventions designed for humans.

Fairness in multi-task learning (MTL) is challenging when heterogeneous output spaces must be controlled through a shared multi-head model. Existing fairness methods are often tied to a single prediction type, while MTL optimizers coordinate task utility without specifying how heterogeneous group disparities should be represented, aggregated, or propagated through task-specific heads. We introduce FairMT, which frames fair heterogeneous MTL through a coordinate-allocation view. FairMT builds a Fairness-Coordinate Interface for binary detection, one-vs-rest multi-class classification, and scalar regression, and instantiates these coordinates asymmetrically using the current best-performing valid group as a detached reference. This directs correction toward groups falling behind the reference, reducing disparities while better preserving task utility. Asymmetric Heterogeneous Fairness Disparity Aggregation (AHFDA) then adaptively allocates constraint pressure over normalized coordinates and converts heterogeneous disparities into a single differentiable constraint signal. Finally, a Head-Aware Optimization Proxy maps the allocated constraint through task-head geometry into task-combination weights for primal--dual training, reducing routing bias caused by heterogeneous head scales, sensitivities, and decision geometries. Experiments on visual and language MTL benchmarks show that FairMT reduces group-level disparities across heterogeneous output types while maintaining utility close to utility-oriented MTL optimizers.

Fair classification faces a twofold challenge: maintaining high classification accuracy while limiting reliance on sensitive information. Our method FairSplit achieves this by learning a decomposition of the embedding space into a Fair Space used for prediction and a Sensitive Space that isolates sensitive information. During training, the model must simultaneously learn the classification task and separate fair from sensitive information. The Sensitive Space gives sensitive attributes a destination, so the Fair Space can be used for prediction without relying on them. This decomposition is guided via a learnable mask and the Hilbert-Schmidt Independence Criterion (HSIC) as a measure of statistical dependence. We provide theoretical guarantees showing that the demographic parity gap is upper-bounded by both (i) the learned dimensionality of the Fair Space and (ii) the HSIC between the Fair Space and sensitive attributes. FairSplit effectively handles multiclass tasks and non-binary sensitive attributes, overcoming key limitations of existing fair classification methods. Experiments on established fairness benchmarks show that FairSplit achieves competitive accuracy-fairness trade-offs.

Identity-preserving text-to-video generation (IPT2V) empowers users to produce diverse and imaginative videos with consistent human facial identity. Despite recent progress, existing methods often suffer from significant identity distortion under large facial pose variations or facial occlusions. In this paper, we propose FaithfulFaces, a pose-faithful facial identity preservation learning framework to improve IPT2V in complex dynamic scenes. The key of FaithfulFaces is a pose-shared identity aligner that refines and aligns facial poses across distinct views via a pose-shared dictionary and a pose variation–identity invariance constraint. By mapping single-view inputs into a global facial pose representation with explicit Euler angle embeddings, FaithfulFaces provides a pose-faithful facial prior that guides generative foundations toward robust identity-preserving generation. In particular, we develop a specialized pipeline to curate a high-quality video dataset featuring substantial facial pose diversity. Extensive experiments demonstrate that FaithfulFaces achieves state-of-the-art performance, maintaining superior identity consistency and structural clarity even as pose changes and occlusions occur. The code and dataset pipeline will be released.

Large Language Models (LLMs) are increasingly used to brainstorm and evaluate research ideas, yet assessing such judgments is fundamentally difficult because the true impact of a new idea may take years to emerge. We address this challenge by using the impact forecasting of human-authored manuscripts as a verifiable proxy task. In a prospective forecasting study, we find that frontier LLMs fail to reliably distinguish high-impact papers from ordinary publications, suggesting that static text-based judging is insufficient for scientific evaluation. To address this limitation, we propose $\textbf{FAME}$ ($\underline{\text{F}}$orecasting $\underline{\text{A}}$cademic Impact via Continuous-Time $\underline{\text{M}}$anifold $\underline{\text{E}}$volution), a spatiotemporal framework for modeling the dynamic trajectories of scientific topics. FAME projects papers into a dynamic latent space informed by textual features and a verified knowledge-flow graph, learning geometric constraints that align impactful manuscripts with the forward momentum of their fields. Experiments on 3,200 arXiv papers across three fast-evolving subfields show that FAME consistently and substantially outperforms state-of-the-art LLM evaluators in prospective multidimensional impact forecasting. Furthermore, integrating FAME's dynamic geometric signals into LLMs significantly improves their forecasting performance. These results support manuscript impact forecasting as a useful, measurable proxy benchmark and position FAME as a strong, trajectory-aware foundation for automated scientific evaluation.

Human-centric video customization, particularly at the garment level, has shown significant commercial value. However, existing approaches cannot support low-latency and interactive garment control, which is crucial for applications such as e-commerce and content creation. This paper studies how to achieve interactive multi-garment video customization while preserving motion coherence using only single-garment video data. We present FashionChameleon, a real-time and interactive framework for human-garment customization in autoregressive video generation, where users can interactively switch garment during generation. FashionChameleon consists of three key techniques: (i) Instead of training on multi-garment video data, we train a Teacher Model with In-Context Learning on a single reference–garment pair. By retaining the image-to-video training paradigm while enforcing a mismatch between the reference and garment image, the model is encouraged to implicitly preserve coherence during single-garment switching. (ii) To achieve consistency and efficiency during generation, we introduce Streaming Distillation with In-Context Learning, which fine-tunes the model with in-context teacher forcing and improves extrapolation consistency via gradient-reweighted distribution matching distillation. (iii) To extend the model for interactive multi-garment video customization, we propose Training-Free KV Cache Rescheduling, which includes garment KV refresh, historical KV withdraw, and reference KV disentangle to achieve garment switching while preserving motion coherence. Our FashionChameleon uniquely supports interactive customization and long-video extrapolation, while achieving real-time generation at 23.8 FPS on a single GPU, 30-180$\times$ faster than existing baselines.

The paradigm of scaling Large Language Models (LLMs) in both parameter size and test time has pushed the boundaries of AI capabilities, but at the cost of making the traditional generative evaluation paradigm prohibitively expensive, therefore making the latency of LLM's in-training downstream performance evaluation unbearable. However, simple metrics like training loss (perplexity) are not always correlated with downstream performance, as sometimes their trends diverge from the actual task outcomes. This dilemma calls for a method that is computationally efficient and sufficiently accurate in measuring model capabilities. To address this challenge, we introduce a new in-training evaluation paradigm that uses a lightweight probe for monitoring downstream performance. The probes take the internal representations of LLM checkpoints (during training) as input and directly predict the checkpoint's performance on downstream tasks measured by \emph{success probability} (i.e., pass@1). We design several probe architectures, validating their effectiveness using the OLMo3-7B's checkpoints across a diverse set of downstream tasks. The probes can accurately predict a checkpoint's performance (with avg. AUROC$>$0.75), have decent generalizability across checkpoints (earlier predicts later), and reduce the computation latency from $\sim$1 hr (using conventional generative evaluation method) to $\sim$3 min. In sum, this work presents a practical and scalable in-training downstream evaluation paradigm, enabling a more agile, informed, and efficient LLM development process.

Diffusion large language models promise parallel token generation, yet inference remains bottlenecked by deciding which masked tokens can be safely committed together. Fast-dLLM addressed this with KV caching and confidence-guided parallel decoding, but its decoding theory uses a homogeneous high-confidence assumption that effectively reduces each candidate set to its weakest selected token. We argue that this leaves speed on the table because real decoding steps exhibit heterogeneous confidence profiles. We propose Fast-dLLM++, a training-free extension that introduces Fr\'{e}chet profile decoding: selecting parallel commit sets from the full sorted confidence profile rather than a single worst-case confidence. The resulting rule is a heterogeneous-confidence generalization of Fast-dLLM's factor selector and it recovers the previous rule exactly in the equal-confidence case and adds a provable heterogeneity bonus when the selected tokens have uneven confidences. Fast-dLLM++ leaves the model, diffusion process, and cache implementation entirely unchanged, making it a drop-in replacement for existing Fast-dLLM decoding. Experiments on GSM8K, MATH, HumanEval, and MBPP with the LLaDA-8B model show that the theoretical improvement translates directly into empirical gains: profile-aware selection improves the accuracy-throughput frontier by exploiting safe parallelism that weakest-token rules miss, achieving up to 37\% higher throughput at comparable accuracy. Our anonymous code release is at https://anonymous.4open.science/r/fast-dllm-plus-plus/.

We introduce FLASH-MAX, a shallow, exact-by-construction neural network architecture for predicting homogeneous electromagnetic fields from sparse pointwise observations. Each hidden neuron represents a separate exact solution to Maxwell's equations, so that the network satisfies the governing equations symbolically by construction and can be trained end-to-end from sparse data within seconds. We prove a universal approximation result showing that this exact model class remains universal on arbitrary domains. FLASH-MAX reaches sub-1% relative validation error from about 1K sparse pointwise observations in seconds, all while maintaining a zero PDE residual, and keeps single digit errors even for only 100 observations sampled from 3D space. These results suggest that moving governing structure from the loss into the hypothesis class can dramatically improve the trade-off between precision and optimization speed in scientific machine learning.

3D semantic occupancy prediction provides dense geometric and semantic scene representations for autonomous driving, where both high accuracy and low response latency are crucial for safe downstream forecasting and planning. Existing methods usually exhibit different accuracy-latency characteristics. Performance-oriented models provide stronger geometric and semantic predictions, while deployment-friendly models provide faster responses. This motivates EvoOcc, a fast-slow evolutionary occupancy prediction framework that jointly exploits their complementary strengths to improve the accuracy-latency balance without changing the upstream architectures. EvoOcc models evolutionary occupancy prediction as a continuous time controlled dynamical system, enabling flexible adaptation to irregular prediction arrival of both fast and slow system. The framework further decomposes the evolution process into dual state evolution where ego state evolution for the deterministic coordinate shifts and scene state evolution for the genuine geometric and semantic scene changes. Extensive experiments on three different fast-slow system pairs and irregular prediction arrivals show that EvoOcc consistently improves over fast system baselines while maintaining much lower latency than the slow system, indicating a favorable accuracy-latency balance and stable behavior under practical inference situations.

Large language model training with sequence-parallel data-parallel (SP-DP) is bottlenecked by the interplay between intra-step activation redistribution and inter-step gradient synchronization. Existing infrequent-communication methods like FedAvg keep states local or reset them lack convergence guarantees and are unstable under long-context sequence-parallel regimes where each device holds only a partial activation shard; Local Adam synchronizes all states jointly and is convergent but triples the collective payload, and its uniform synchronization period cannot distinguish the semantically mandatory intra-step All-to-All from the temporally deferrable inter-step AllReduce. We propose FAUST (Federated Asynchronous Update with Staggered Timescales), a compiler-orchestrated training system that composes compile-time sequence-parallel graph rewriting with tiered runtime synchronization, assigning independent averaging periods to parameters ($K_x$), first moments ($K_u = 3K_x$), and second moments ($K_v = 6K_x$) according to their impulse-response half-lives, while preserving convergence under the orthogonal collective constraint that intra-step All-to-All must retire before inter-step AllReduce begins. Our analysis shows that while first-moment drift dominates the convergence rate in-distribution, high-fidelity convergence guarantees require at least periodic synchronization of the second moment. Experiments on language models up to 1.3B parameters show that FAUST incurs $303{\times}$ less wall-clock communication overhead than standard DDP and $1.78{\times}$ less than the previous state-of-the-art DES-LOC, while achieving perplexity within $4.6\%$ of the fully-synchronized baseline in $27\%$ less wall-clock time. On bandwidth-constrained links, FAUST delivers $48$--$303{\times}$ speedup over DDP.

Diffusion models generate data through a continuum of denoising problems, and are widely observed to reveal coarse structure before fine detail. Yet, this intuition is mostly empirical and qualitative. We introduce \emph{feature information dynamics}, an information-theoretic framework for localizing when a feature is generated during diffusion. Using the I-MMSE identity, we connect the rate of feature mutual information change to a gap between optimal unconditional and feature-conditional denoising losses, yielding practical estimators for feature information density. We further develop a chained decomposition that separates shared from incremental information in a feature hierarchy. We use this framework first to quantitatively confirm spectral autoregression in pixel diffusion, and then to extend the analysis beyond frequency: under a class$\to$mask$\to$Canny conditioning chain, the per-feature information densities differ across pixel, SDVAE, VAVAE, and RAE, exposing fundamental differences between these representations.

Large language models (LLMs) are increasingly accessed as remotely hosted services by edge and enterprise clients that cannot run frontier models locally. Since models vary widely in capability and price, routing queries to models that balance quality and inference cost is essential. Existing router approaches assume access to centralized query-model evaluation data. However, these data are often fragmented across clients, such as end users and organizations, and are privacy-sensitive, which makes centralizing data infeasible. Additionally, per-client router training is ineffective since local evaluation data is limited and covers only a restricted query distribution and a biased subset of model evaluations. We introduce the first federated framework for LLM routing, enabling clients to learn a shared routing policy from local offline query-model evaluation data. Our framework supports both parametric multilayer perceptron router and nonparametric K-means router under heterogeneous client query distributions and non-uniform model coverage. Across two benchmarks, federated collaboration improves the accuracy-cost frontier over client-local routers, both via increased effective model coverage and better query generalization. Our theoretical results also validate that federated training reduces routing suboptimality.

Low-Rank Adaptation (LoRA) has become ubiquitous for efficiently fine-tuning foundation models. However, federated fine-tuning using LoRA is challenging due to suboptimal updates arising from traditional federated averaging of individual adapters. Existing solutions either incur prohibitively high communication cost that scales linearly with the number of clients or suffer from performance degradation due to limited expressivity. We introduce Fed-SB, a novel approach for federated fine-tuning of LLMs using LoRA-SB, a recently proposed low-rank adaptation method. LoRA-SB optimally aligns the optimization trajectory with the ideal low-rank full fine-tuning projection by learning a small square matrix (R) between adapters B and A, keeping other components fixed. Direct averaging of R guarantees exact updates, substantially reducing communication cost, which remains independent of the number of clients, and enables scalability. Fed-SB achieves state-of-the-art performance across commonsense reasoning, arithmetic reasoning, and language inference tasks while reducing communication costs by up to 230x. In private settings, Fed-SB further improves performance by (1) reducing trainable parameters, thereby lowering the noise required for differential privacy and (2) avoiding noise amplification introduced by other methods. Overall, Fed-SB offers a state-of-the-art, efficient, and scalable solution for both private and non-private federated fine-tuning. Our code is available publicly at: https://github.com/CERT-Lab/fed-sb.

In real-world deployment, large language models (LLMs) are frequently updated through post-training techniques to maintain up-to-date knowledge. Yet their reliability depends not only on what the LLM knows, but also on whether it knows what it knows—a self-assessment known as the feeling of knowing (FoK). FoK is the signal behind selective generation, retrieval triggering, and model routing; when miscalibrated, it leads systems to overconfidence on unknown questions or to refuse ones they could have answered. As LLMs are updated through posttraining, however, we discover that what an LLM knows can change without a corresponding update to its FoK. This creates a knowledge–FoK desynchronization challenge, where FoK estimators calibrated to the pre-trained LLM become unreliable as the model’s knowledge state evolves. To address this challenge, we propose a two-channel design: a knowledge channel that changes what the LLM knows and a FoK channel that judges whether the current knowledge state supports answering. We train these channels with a three-stage procedure: onpolicy supervised fine-tuning (OP-SFT) updates the knowledge channel from corrected on-policy answers, knowledge-delta FoK training (K∆-FT) trains the FoK channel over intermediate knowledge states, and FoK-guided policy optimization (FGPO) uses the judgment to generate the corresponding answer or explanation. Experiments on three datasets show that explicit FoK alignment substantially improves judgment quality on updated knowledge. The code is available at https: //anonymous.4open.science/status/Know-thyself-code-8D65.

Benchmark quality is critical for meaningful evaluation and sustained progress in time series forecasting, particularly with the rise of time series foundation models (TSFMs). Existing benchmarks often have limited domain coverage or overlook real-world settings such as tasks with covariates. Their aggregation procedures frequently lack statistical rigor, making it unclear whether performance differences reflect true improvements or random variation. Many benchmarks lack consistent evaluation infrastructure or are too rigid for integration into existing pipelines. To address these gaps, we propose fev-bench, a benchmark of 100 forecasting tasks across seven domains, including 46 with covariates. Supporting the benchmark, we introduce fev, a lightweight Python library for forecasting evaluation emphasizing reproducibility and integration with existing workflows. Using fev, fev-bench employs principled aggregation with bootstrapped confidence intervals to report performance along two dimensions: win rates and skill scores. We evaluate various models on fev-bench and find that existing TSFMs often miss accuracy by ignoring covariates, pointing to a clear direction for future work.

Visual grounding aims to localize image regions corresponding to natural language expressions. While recent Large Vision-Language Models (LVLMs) have shown impressive multi-modal understanding capabilities, their application to visual grounding typically requires fine-tuning and architectural modifications. This requirement, however, can be ignored, considering that text and images tend to have similar feature representations that appear to be approximately linearly disentangled, enabling cleaner extraction of spatial information from LVLMs without any task-specific training. From this viewpoint, we propose an attention-head discovery framework that requires zero labeled grounding samples and no architectural modifications, and identifies discriminative localization heads without manual inspection. Through dual prompting with target and contrastive descriptions, we compute differential residual representations and project them through attention head output matrices to measure per-head spatial contributions via four complementary scores. By aggregating signals using importance-weighted query difference scores from only the top-10 attention heads, we outperform training-free non-LVLM baseline by up to 27.95% on RefCOCO, 21.93% on RefCOCO+, and 8.40% on RefCOCOg. Our method outperforms LVLM baseline by up to 8.04% on RefCOCO without requiring ground-truth category labels.

Causal discovery often starts from a set of variables that were measured together because they are believed to describe related parts of the same underlying process. Standard continuous DAG learners encode data fit, sparsity, and acyclicity, but they can still return graphs that fragment into isolated variables or disconnected components. We study a simple structural prior for this setting: the learned causal DAG should remain sparse while its undirected skeleton should be weakly connected. We introduce Fiedler regularization, a differentiable spectral penalty based on the algebraic connectivity of a smooth support-preserving skeleton of the learned weighted adjacency matrix. The resulting penalty can be added modularly to differentiable causal discovery objectives, providing connectivity control without changing the underlying learner. We instantiate the approach for GOLEM, graph autoencoder, and DAG-GNN models. Across sparse connected synthetic DAGs up to 200 variables, including connected-conditioned Erdős-Rényi, sparse connected, scale-free, and small-world graph families, Fiedler-regularized learners improve fragmentation control and structural recovery across learner families. These results position algebraic connectivity as a principled and practical prior for learning sparse causal DAGs in settings where the measured variables are expected to have a connected causal structure.

Time series forecasting is an actively researched problem with diverse applications. Recent work has shown that simple linear models can compete with complex deep learning architectures in terms of forecasting accuracy. This has led to the design of several small models, yet the reasons behind their effectiveness remain insufficiently understood. We address this gap through a theoretical framework grounded in reduced-rank regression and kernel analysis. We show both analytically and empirically that the intrinsic complexity of many TSF tasks is lower than commonly assumed and that the performance of linear models is linked to their alignment with low-rank data structures. Building on these insights, we propose a new design for low-rank neural networks that incur lower computational cost than linear models. Specifically, we introduce a simple filter bank architecture that provides a principled way to control model complexity through tunable and interpretable hyperparameters that directly correspond to the rank of the forecasting model. The proposed architecture also serves as a versatile building block for constructing deeper networks. Our evaluation shows that filter bank architectures achieve state-of-the-art results on most long-term TSF benchmarks, with lower computational cost than recently proposed solutions.

We present a vision–language pretraining (VLP) framework for the medical domain that focuses on preserving fine-detail clinical information, such as disease severity, which plays a critical role in downstream medical applications. Despite recent advances in medical VLP, existing approaches often struggle to retain such subtle semantics. This limitation mainly stems from the use of contrastive learning schemes that treat supervision as hard one-hot labels, overlooking the ordinal and overlapping nature of clinical findings and thereby introducing false negatives. To overcome these challenges, we propose FinAl, a fine-grained alignment approach that incorporates soft alignment targets into vision-language pretraining. These targets are derived from structured radiology reports and encode relative semantic relationships among clinical descriptions, allowing fine-grained linguistic information to be transferred into the visual representation space. Extensive experiments show that FinAl improves performance on fine-grained clinical tasks, including severity grading and severity-aware retrieval, with additional evaluations on challenging fine-grained scenarios further supporting its robustness. At the same time, the proposed method maintains competitive performance on standard coarse-grained clinical benchmarks.

Selecting a finite dictionary of observables whose span is Koopman-invariant is a central challenge in data-driven Koopman operator approximation. We address this problem by exploiting zero-block structure in Extended Dynamic Mode Decomposition (EDMD) matrices. We show that any sub-dictionary whose span is Koopman-invariant induces an exact zero block in the EDMD matrix, even for finite data. We then show that such blocks can be detected by applying PageRank to a row-normalized EDMD matrix constructed from a large initial dictionary. The theory extends to approximately invariant subspaces and yields stronger guarantees for personalized PageRank (PPR) when the seed observables lie inside the target block and reach all observables in that block. Combining EDMD concentration bounds with PageRank perturbation theory gives end-to-end detection guarantees with $O(M^{-1/2})$ finite-sample scaling and explicit constants. More generally, without assuming an invariant subspace exists, high PPR mass on a sub-dictionary controls discounted multi-step leakage from the seed observables. Numerical experiments on the Duffing oscillator, Van der Pol oscillator, Lorenz system, and a three-well Ramachandran potential suggest that the method identifies compact, interpretable dictionaries with accurate predictions.

Vision-Language-Action (VLA) models have advanced rapidly by leveraging large-scale vision-language priors and robot trajectory data, yet they remain weak at following fine-grained human instructions. We argue that this limitation stems from a fundamental mismatch between language and action supervision in existing robot datasets. Most open-source datasets annotate each trajectory with a goal-oriented task description, while leaving the underlying execution process largely unspecified, including motion trajectories, contact and approach patterns, object state transitions, recovery behavior, and final configurations. Since different choices along these dimensions can lead to substantially different action sequences, coarse task descriptions induce a many-to-one mapping from action trajectories to language, preventing precise action-instruction alignment. To address this problem, we propose FineVLA, a framework for constructing and leveraging fine-grained embodied action instructions for VLA learning. FineVLA includes FineVLA-Tool, a scalable pipeline for data cleaning, temporal alignment, and fine-grained annotation; RoboFine-VLM, a vision-language model for fine-grained robotic action understanding and annotation; RoboFine-Bench, a benchmark for evaluating fine-grained robotic action understanding via VQA and captioning; and FineVLA-Policy, a VLA policy trained with fine-grained instructions on open-source ALOHA robot data. Across benchmark evaluation and policy learning experiments, FineVLA demonstrates that high-density, action-aligned language supervision leads to more controllable and instruction-sensitive robot policies. Our results highlight fine-grained instruction alignment as a critical step toward VLA systems that not only complete tasks, but complete them in the way humans specify.

We study token dynamics in deep Mamba models, with a particular focus on the attracting role of the first token. Unlike attention mechanisms, where token interactions are governed by positive scalar coefficients, Mamba dynamics involve full matrix-valued interaction coefficients with no positivity or definiteness guarantees. This structural difference makes Mamba dynamics substantially harder to analyze and prevents the direct application of existing techniques for attention-based models. To overcome this challenge, we introduce a new analytical framework based on a high-dimensional invariant box and a sharp exit-time argument, which together recover an effective form of sign-definiteness. Using this framework, we characterize the possible limit points of token trajectories and prove explicit exponential convergence rates. Our analysis reveals that the first token has privileged dynamics: its limiting direction attracts subsequent tokens, inducing token clustering and attention-sink phenomena in deep Mamba models. We further show that this attraction mechanism extends to broader sequence dynamics, including state-space models, causal attention, and classical time-series models. Experiments on Falcon-Mamba-7B validate our predictions, showing that concentration on the first token increases with depth and dominates in later layers.

Recent advances in flow-based offline reinforcement learning (RL) have achieved strong performance by parameterizing policies via flow matching. However, they still face critical trade-offs among expressiveness, optimality, and efficiency. In particular, existing flow policies interpret the $L_2$ regularization as an upper bound of the 2-Wasserstein distance ($W_2$), which can be problematic in offline settings. This issue stems from a fundamental geometric mismatch: the behavioral policy manifold is inherently anisotropic, whereas the $L_2$ (or upper bound of $W_2$) regularization is isotropic and density-insensitive, leading to systematically misaligned optimization directions. To address this, we revisit offline RL from a geometric perspective and show that policy refinement can be formulated as a local transport map—an initial flow policy augmented by a residual displacement. By analyzing the induced density transformation, we derive a local quadratic approximation of the KL-constrained objective governed by the Fisher information matrix, enabling a tractable anisotropic optimization formulation. By leveraging the score function embedded in the flow velocity, we obtain a corresponding quadratic constraint for efficient optimization. Our results reveal that the optimality gap in prior methods arises from their isotropic approximation. In contrast, our framework achieves a controllable approximation error within a provable neighborhood of the optimal solution. Extensive experiments demonstrate state-of-the-art performance across diverse offline RL benchmarks.

Spiking Neural Networks (SNNs) are a promising framework for event-driven temporal processing. Prior work has improved temporal modeling through richer neuron dynamics and network-level mechanisms such as recurrence and delays, but it remains unclear how individual spiking neurons should specialize within a network. In this work, we introduce \textbf{FiTS}, a spiking neuron that factorizes temporal computation within each neuron into Frequency Selectivity (FS) and Temporal Shaping (TS). The FS module parameterizes each neuron's target frequency as the maximizer of its subthreshold magnitude response, while the TS module reshapes when frequency components contribute to membrane voltage accumulation through group-delay modulation. On auditory benchmarks where frequency selectivity and timing are central to the input structure, FiTS consistently improves over a plain Leaky Integrate-and-Fire (LIF) baseline in simple feedforward SNNs without recurrence or network-level delays, while remaining competitive with strong temporal SNN baselines. Beyond accuracy, the learned target frequencies and group-delay shifts provide interpretable neuron-level summaries of the frequency and timing organization learned within the network.

Generative models such as diffusion and flow matching have become dominant paradigms for visuomotor policy learning, yet their reliance on iterative denoising incurs high inference latency incompatible with real-time robotic control. We present a Fast Legendre-polynomial Action policy via Sparse History-anchored flow (FLASH), which replaces discrete action-chunk generation with continuous Legendre polynomial trajectory representation. Specifically, by fitting expert demonstrations under sparse temporal sampling, FLASH enables a single inference to cover a significantly extended action horizon. To further accelerate generation, FLASH initiates the flow matching process from history polynomial coefficients rather than uninformative Gaussian noise, shortening the transport distance and enabling accurate single-step inference. Moreover, analytic polynomial differentiation directly provides desired velocity feed-forward signals to the torque controller without numerical approximation. Extensive experiments on five simulated and two real-world manipulation tasks demonstrate that FLASH achieves state-of-the-art success rates ($\ge 92\%$ across all tasks), a per-episode inference time of $31.40\,ms$ (up to $175\times$ faster than diffusion policies and $18\times$ faster than prior flow matching policies), up to $4\times$ faster training convergence than ACT, and $5\times$ to $7\times$ reduction in controller tracking error compared to discrete-action baselines.

Flow matching models have emerged as a dominant paradigm for generative posterior sampling in imaging inverse problems. Existing solvers follow two distinct lines of work: trajectory-space methods, which refine sampling dynamics from random initial noise, and source-space methods, which optimize the initial noise itself. While prior work treats these as independent design choices, we show that they are fundamentally coupled. We derive a unified reconstruction error bound that links initialization proximity and cumulative sampling stochasticity through a multiplicative envelope, yielding two key insights. First, we identify a sufficiency regime in which a moderately optimized proximal source initialization is enough to attain the optimal error floor, rendering exhaustive source-space optimization unnecessary. Second, this initial advantage is fragile: it is eroded by excessive stochasticity accumulation during guided sampling, necessitating a mechanism that bounds the cumulative noise injection. These findings motivate FLINT, a two-stage framework that operationalizes this coupling. FLINT establishes a high-fidelity starting point via a manifold-constrained proximal objective in only a few iterations, then performs guided sampling under a bounded stochasticity schedule that prevents erasure of the initialization advantage. Across diverse inverse problems, FLINT delivers state of the art perceptual quality and competitive distortion metrics while maintaining superior computational efficiency.

Sharpness-Aware Minimization (SAM) improves generalization in continuous deep learning, yet applying it to binary-weight networks via latent-space heuristics suffers from a fundamental geometric mismatch: continuous perturbations do not faithfully probe the discrete loss landscape on $\{-1,+1\}^n$. We introduce BOLD-SAM, the first sharpness-aware optimizer that operates natively on the Boolean hypercube, replacing the Euclidean $\ell_p$-ball with a $k$-bit Hamming ball and solving the resulting discrete min-max problem via greedy ascent followed by sharpness-aware bit-flip descent. The objective is theoretically justified from three complementary perspectives: PAC-Bayes, compression, and distributionally robust optimization. We prove an approximation guarantee for the ascent step and a finite-time convergence bound for the descent, both governed by the discrete interaction Hessian, which emerges as the unifying quantity linking ascent quality to convergence rate. We further connect discrete flatness to Forman-Ricci curvature and algorithmic complexity. Experiments on a wide range of architectures and datasets demonstrate consistent improvements in clean accuracy and out-of-distribution robustness over standard binary training and latent-space SAM baselines.

Many multimodal alignment problems require mapping a source modality into a fixed space defined by a target modality, where the target space is not merely an intermediate embedding but is used directly for downstream analysis. In this setting, successful alignment should preserve both the global structure of the target distribution and the instance-level correspondence observed in paired source--target training data. Existing objectives typically address only part of this goal: contrastive losses encourage relative similarity, distribution-matching losses improve marginal agreement, and pointwise regression fits each source sample to its paired target without explicitly modeling target-space path geometry. We propose FlowAlign, a flow-guided target-space alignment method that repurposes conditional flow matching as a training-time alignment signal rather than an inference-time transport model. FlowAlign uses the learned flow field to guide the alignment of source-side representations with the fixed target space, encouraging them to preserve paired correspondence while remaining compatible with the global target-space structure. We further provide theoretical analysis showing how FlowAlign helps improve paired correspondence by controlling endpoint accuracy and capturing local target-space geometry. Experiments on a controlled synthetic task and paired single-cell multiome benchmarks show that FlowAlign improves alignment performance while preserving target-space structure for downstream analysis.

Automated multi-agent systems offer clear advantages over manually designed ones in scalability and adaptability, but existing workflow topology methods still face important limitations. Search-based methods are often computationally expensive, textual-gradient-based methods rely on coarse-grained feedback, and existing generation-based methods are not well suited to discrete workflow topologies with complex dependencies. To address these limitations, we propose FlowMAS, a multi-agent workflow topology method based on Generative Flow Networks (GFlowNets). FlowMAS models workflow generation as reward-guided flow over the topology space and introduces three components: a GFlowNet-based topology generation backbone, a curiosity-driven module for structure-aware exploration, and an information-guided optimization module for evaluating intermediate topologies. Concretely, the curiosity-driven module encourages exploration of structurally novel workflows, while the information-guided module measures both the information contribution and the communication efficiency of different operators to favor more informative and effective collaboration patterns. Experiments on six benchmark datasets with three LLM backbones show that FlowMAS consistently outperforms multiple baselines.

Balancing policy expressiveness with the exploration-exploitation trade-off is a core challenge in online Reinforcement Learning (RL). While Stochastic Differential Equation (SDE)-based diffusion policies can represent complex, multimodal action distributions, they suffer from two critical limitations: their stochastic reverse processes render entropy intractable (necessitating heuristic exploration), and computing policy gradients through long denoising chains is expensive and unstable. In this work, we show that ODE-based flow matching inherently resolves these issues by enabling both simulation-free policy optimization and tractable entropy computation. Building on this, we introduce Flow Matching Policy Optimization with Mirror Descent and Entropy Constraints (FMER). Our framework exploits this insight in three ways. First, we theoretically establish that minimizing an advantage-weighted conditional flow matching loss acts as a simulation-free surrogate for policy mirror descent. This steers the velocity field toward high-value regions while entirely avoiding backpropagation through the ODE solver. Second, we derive an analytic entropy objective that corrects for the density distortion caused by the $\tanh$ transformation (mapping an unbounded latent space to bounded actions), thereby facilitating principled maximum-entropy optimization. Finally, we dynamically tune the mirror descent temperature based on the effective sample size to enforce a robust trust region during training. Empirical evaluations demonstrate that FMER achieves superior performance on the challenging sparse-reward FrankaKitchen environment, while maintaining competitive results across standard dense-reward MuJoCo benchmarks. Code is available at \url{https://anonymous.4open.science/r/submission-25E2/}.

In recent years, agentic workflows have been widely applied to solve complex human tasks. However, existing workflow construction still faces key challenges, including human-dependent workflow construction, the lack of graph-level execution feedback, and the inability to repair errors in-loop during long-horizon construction. To address these challenges, we propose FlowSteer, a new paradigm of Agent Designing Agentic Workflows — a single agent itself end-to-end designs the workflow that a downstream executor runs. To support this paradigm, we introduce the Workflow Canvas, a novel executable graph-state environment that returns syntax-checked execution feedback for every atomic edit. Built on the canvas, we further propose Reinforced Progressive Canvas Editing, in which a lightweight policy agent issues one atomic edit per turn conditioned on real canvas feedback, and is trained end-to-end via reinforcement learning. Moreover, FlowSteer provides a plug-and-play framework that supports diverse operator libraries and interchangeable LLM backends. Experimental results on twelve datasets show that FlowSteer significantly outperforms baselines across various tasks. Our code is available at https://anonymous.4open.science/r/FlowSteer-9B2E.

Ground-to-space astronomical super-resolution requires recovering space-quality images from ground-based observations that are simultaneously limited by pixel sampling resolution and atmospheric seeing, which imposes a stochastic, spatially varying PSF that cannot be resolved through upsampling alone. Existing methods rely on synthetic training pairs that fail to capture real atmospheric statistics and are prone to either over-smoothed reconstructions or hallucination sources with no physical counterpart in the observed sky. We propose FluxFlow, a conservative pixel-space flow-matching framework that incorporates observation uncertainty and source-region importance weights during training, and a training-free Wiener-regularised test-time correction to suppress hallucination sources while preserving recovered detail. We further construct the DESI--HST Dataset, the large-scale real-world benchmark comprising 19,500 real co-registered ground-to-space image pairs with real atmospheric PSF variation. Experiments demonstrate that FluxFlow consistently outperforms existing baseline methods in both photometric and scientific accuracy.

Verifying the local robustness of neural networks under $\ell_0$ perturbations is inherently combinatorial: a verifier must reason over all sparse subsets of perturbed pixels. This paper presents FocusBranch, a combinatorial branch-and-bound framework for $\ell_0$ robustness verification. The key insight is to decompose the $\ell_0$-ball using nodes, where each node consists of a focus pixel set $\mathcal{S}$ and a level $l$, requiring at least $l$ perturbed pixels to lie in $\mathcal{S}$. This abstraction unifies two branching mechanisms: concrete nodes branch through upper shadows to fix additional perturbed locations, while super nodes branch through covering designs to cover many perturbation cases with fewer child nodes. FocusBranch proves node safety by concretizing linear relaxations over node-constrained $\ell_0$-balls and recursively branches on unknown nodes. Rather than relying on a fixed verification strategy, it samples nodes across levels to estimate success rates and costs, then selects a starting level and branching strategy predicted to minimize verification time. Experiments on fully connected and convolutional classifiers for MNIST, Fashion-MNIST, and CIFAR-10 show that FocusBranch verifies $\ell_0$ local robustness effectively across architectures and outperforms state-of-the-art verifiers.

Contact-rich manipulation requires precise control and complex physical interaction with the environment, posing significant challenges for robot policy learning. Vision-only policies struggle with such tasks, as contact states and force feedback are difficult to infer from vision alone. Existing methods mitigate this issue by incorporating force or tactile sensing. However, they often treat these signals as auxiliary observations rather than using them to model future interaction dynamics, leading to inefficient modality utilization and limited generalization. To address this, we propose ForceLDM, a force-aware latent dynamics model that enables the policy to anticipate future contact and motion dynamics in latent space and use them to guide action generation. Rather than predicting future observations in pixel space, which often requires large amounts of data and risk overfitting irrelevant details, ForceLDM learns a compact task-relevant representation through knowledge distillation. Specifically, a teacher network with privileged access to future force/torque and optical flow signals extracts future dynamic features, which are then distilled into a student network trained only on current observations. This enables the student to reason about upcoming contacts and physical dynamics at deployment without relying on future sensory inputs. To improve robustness and generalization, we further introduce a curriculum-based progressive noise injection strategy to mitigate over-reliance on future features. Experiments on five real-world contact-rich manipulation tasks demonstrate that ForceLDM significantly outperforms state-of-the-art baselines, generalizes effectively to novel objects, and remains robust to environmental perturbations.

Progress in language model development is often driven by comparative decisions: which architecture to adopt, which pretraining corpus to use, or which training recipe to apply. Making these decisions well requires reliable performance forecasts, yet the two commonly used signals are fundamentally limited. Cross-entropy loss is poorly aligned with downstream capabilities, and direct downstream evaluation is expensive, sparse, and often uninformative at early training stages. Instead, we propose to construct proxy metrics by aggregating token-level statistics, such as entropy, top-k accuracy, and expert token rank, from a candidate model's next token distribution over expert-written solutions. Across three settings, our proxies consistently outperform loss- and compute-based baselines: 1) For cross-family model selection, they rank a heterogeneous population of reasoning models with mean Spearman $\rho = 0.81$ (vs. $\rho = 0.36$ for cross-entropy loss); 2) For pretraining data selection, they reliably rank 25 candidate corpora for a target model at roughly $10{,}000\times$ less compute than direct evaluation, pushing the Pareto frontier beyond existing methods; and 3) for training-time forecasting, they extrapolate downstream accuracy across an $18\times$ compute horizon with roughly half the error of existing alternatives. Together, these results suggest that expert trajectories are a broadly useful source of signal for assessing model capabilities, enabling reliable performance forecasting throughout the model development lifecycle.

Large language models (LLMs) have demonstrated strong capabilities in question answering, yet they still frequently suffer from hallucinations on knowledge-intensive tasks. Knowledge graphs (KGs) provide LLMs with structured, interpretable, and updatable factual grounding, making them a promising external knowledge source for reliable reasoning. However, existing LLM-guided graph reasoning methods typically rely on hop-wise greedy or beam-style pruning during evidence retrieval. Such local decision processes are inherently myopic: evidence that appears weak near the source may become crucial only after deeper graph context is explored, causing answer-critical branches to be discarded prematurely and making the reasoning chain difficult to recover. To address this limitation, we propose Foresight-over-Graph (FoG), a foresight-aware evidence retrieval framework for knowledge base question answering (KBQA). FoG iteratively constructs a question-relevant evidence subgraph and uses far-to-near feedback to guide path exploration, and maintains a compact memory subgraph to support continued exploration. Extensive experiments on widely used KBQA benchmarks demonstrate that FoG achieves state-of-the-art performance, with a particularly large improvement of 16.58\% in Hit on CWQ, while also reducing LLM calls and token usage. Our code is available at \url{https://anonymous.4open.science/r/FoG-7273} .

Machine unlearning aims to remove the influence of selected training examples without full retraining. Standard evaluations often summarize unlearning quality with aggregate metrics, such as accuracy- and forgetting-based scores, which can hide localized failures. We study this failure mode at the example level by comparing the predictions of an unlearned model to those of the model retrained after deletion. We show that this pointwise discrepancy can be highly non-uniform: for gradient-ascent and random-labeling methods, with and without retain-set fine-tuning, it grows with geometric proximity to the forget set. We call this phenomenon localized collateral forgetting. Our analysis identifies a mechanism behind the effect: surrogate targets used during unlearning can be inconsistent with the local prediction structure induced by retraining, and this inconsistency propagates through shared representations to nearby examples. Motivated by this mechanism, we propose Local Teacher Distillation, a simple mitigation strategy that replaces random targets with soft labels from a small teacher trained only on retained neighbors of the forget set. On CIFAR-100 partial-class deletion, this local teacher brings the unlearned model substantially closer to retraining, especially near the forget set, while maintaining competitive aggregate unlearning metrics.

Symmetry-aware learning is typically framed as equivariance to a single group action, but many real-world distribution shifts are typed, partial/non-invertible, and compositional (e.g., modality changes, occlusion, subsampling, intervention chains). These shifts are better modeled by a category of transformations whose objects represent data contexts and whose morphisms represent admissible transformations between contexts. We thus extend symmetry-aware machine learning from groups to such transformation categories; enforcing categorical equivariance on neural architectures gives stronger robustness than group equivariance. We ground this in both theory and experiments. We formulate a universal approximation theorem for category-equivariant architectures and show density in the space of equivariant continuous transformations. We present experiments on compositional OOD shifts, demonstrating that enforcing categorical equivariance yields measurable robustness gains over group-equivariant and non-equivariant baselines.

Visual state space models (SSMs) have emerged as efficient alternatives to attention-based backbones, yet their performance remains sensitive to how visual evidence is sampled and propagated. Deformable visual Mamba methods address fixed scanning by predicting adaptive sampling locations, but offset prediction is still driven primarily by local responses, producing locally plausible samples that are weakly coupled to the SSM's recurrent state evolution, particularly in off-center or multi-object scenes. We present Foveal-Mamba, a scan-path-aware deformable visual Mamba framework that reformulates offset prediction as recurrent scan-path prediction. Two components drive this design. First, an Inside-Out Ring Scan organizes visual tokens along a center-to-periphery path, providing a stable propagation prior that does not assume object centering. Second, a lightweight Mamba module embedded within the offset prediction network forms the Recurrent Scan-Path Offset Prediction Network (RS-OPN), which predicts offsets from hidden states accumulated along the ring path. RS-OPN tightly couples deformable sampling with recurrent state evolution and produces coherent, object-aware sampling trajectories. Foveal-Mamba consistently outperforms strong baselines across multiple benchmarks. On ImageNet-1K, Foveal-Mamba-Tiny achieves 84.3\% Top-1 accuracy, surpassing DAMamba-Tiny by 0.5\% with fewer parameters and FLOPs. On ADE20K semantic segmentation, Tiny/Small variants reach 50.8/51.8 mIoU. On COCO, it achieves 48.9/50.2 box AP and 43.7/44.9 mask AP for object detection and instance segmentation, respectively. Qualitative results further demonstrate earlier object focus and more stable activations in multi-object scenes.

In real-world tasks such as robotic control and autonomous driving, agents often face non-Markovian environments due to incomplete observations. Solving such information-impoverished decision tasks requires reconstructing the context from history. However, existing recurrent encoders have two fundamental limitations: First, the model suffers from a Markovian compression bottleneck, i.e., it only processes observation history as input but neglects thought process history. This leads to the loss of deep semantic cues from early interactions, which are easily overwritten by noise over time due to recursion. Second, it lacks the adaptive capability to adjust the memory span for different tasks. To address these limitations, we propose FracEncoder, a context encoder based on fractional-order neural differential equations. First, FracEncoder leverages the nonlocal property of the fractional-order derivative and uses a power-law kernel to directly incorporate the trajectory of the thought process into the evolution of the current state, effectively overcoming the temporal locality limitations of traditional models. Second, the fractional order $\alpha$ is set as an end-to-end learnable parameter. It explicitly characterizes how strongly the system anchors to its history, so the model can discover a suitable memory span for each task. We evaluate FracEncoder on more than ten benchmarks across three settings, namely partially observable tasks, meta reinforcement learning, and delayed-observation control. FracEncoder consistently matches or outperforms representative baselines, including GPIDE, GRU-ODE, ReSeL, and LRU. On the challenging 8-step delayed-observation task, it reaches average returns of $-511.5 \pm 7.9$ on Pendulum and $732.1 \pm 152.7$ on Hopper. As a plug-in module, FracEncoder also yields stable gains across different reinforcement learning algorithms. Ablation studies further show that the learned $\alpha$ aligns with the intrinsic memory demand of each task, providing an interpretable physical measure of the cognitive complexity of non-Markovian tasks. The code is available at https://anonymous.4open.science/r/FracEncoder-60BA.

Streaming Visual Geometry Transformers such as StreamVGGT enable strong online 3D perception, but their KV-cache grows unbounded over long streams, limiting practical deployment. We study bounded-memory streaming geometry from the perspective of memory organization: unlike language modeling, where useful information can often be compressed at token level, geometry-driven inference relies on coherent and mutually compatible observations across views. Under fixed memory budgets, retaining history as isolated entries can progressively fragment the geometric context needed for stable long-horizon matching and fusion. We therefore propose \textbf{FrameVGGT}, a bounded-memory framework that maintains a fixed-capacity set of complementary memory units for streaming geometry. In our implementation, each unit is instantiated as a frame-wise KV segment summarized by a compact key-space prototype, together with a sparse anchor tier for persistent long-range references. Across long-sequence 3D reconstruction, video depth estimation, and camera pose estimation, FrameVGGT achieves favorable accuracy--memory trade-offs under bounded budgets while maintaining more stable geometry over long streams.

Fine-tuning large-scale foundation models across resource-constrained, distributed clients presents significant computational and communication challenges. While merging Low-Rank Adaptation (LoRA) with Federated Learning (FL) enables rank-based flexibility, the interdependence of the adaptation matrices ($A$ and $B$) often complicates optimization. In this work, we theoretically demonstrate that LoRA fine-tuning with a frozen A matrix acts as a stochastic approximation of full fine-tuning, where batch gradients are subject to a random, rank-dependent perturbation. Motivated by this insight, we propose freezing and periodically resampling $A$ matrices at the client level. This strategy decouples the $A$ and $B$ gradients, diminishes LoRA low-rank constraints and reduces both computational and uplink communication overheads by 50%, allowing for the reallocation of resources toward higher adaptation ranks. Furthermore, we show that in this frozen-$A$ regime, the adaptation rank $r$ directly governs the signal-to-noise ratio (SNR) of the perturbed gradients. Specifically, we find that the SNR increases linearly with $r$, implying that clients’ learning rates must be dynamically scaled with their adaptation ranks for higher utility. Building on these foundations, we introduce Frank-LoRA, a federated rank-aware fine-tuning strategy that utilizes frozen, resampled $A$ matrices and rank-dependent learning rates for clients. We also prove the convergence of the algorithm in FL settings with heterogeneous clients resources. Our experiments across benchmark datasets demonstrate that Frank-LoRA consistently outperforms state-of-the-art baselines with half the overhead on clients.

Flow matching has emerged as a standard paradigm for robotic manipulation owing to its strong expressive power for modelling complex, multimodal action distributions, alongside similar approaches like diffusion policy. However, existing methods rely on discretized action chunks, making them brittle to demonstrations collected at heterogeneous control frequencies and prone to temporally inconsistent actions that degrade control stability. In this paper, we propose Frequency-Aware Flow Matching (FAFM), which outputs continuous, temporally consistent actions. To handle heterogeneous frequency input, we transform discrete action sequences into the frequency domain with the discrete cosine transform (DCT), perform flow matching over the resulting coefficients, and reconstruct continuous actions via cosine basis expansion. To generate temporally consistent actions, we regularize the first-order temporal derivative to promote smooth actions. This corresponds to a Sobolev-type constraint that suppresses high-frequency errors and discourages abrupt action changes. Our FAFM is simple, introduces no additional network parameters and applies to standalone flow-matching policies and vision-language action models. Across synthetic toy benchmark, obstacle avoidance, LapGym, and LIBERO, FAFM improves success rates, multimodal expressivity, motion smoothness, convergence speed, robustness to mechanical bias and mixed-frequency input. These gains are consistent when deployed on a real-world Franka robot. Code available at \url{https://anonymous.4open.science/r/FAFM}.

Asynchronous pipeline parallelism can accelerate distributed model training, but suffers from instability caused by inconsistent gradient computations. We identify a fundamental source of this inconsistency: error signals are coupled to forward-time features, not just parameter states. This means that parameters may evolve as long as they preserve forward-equivalence. Leveraging this insight, we introduce FRESCO, which frames asynchronous updates as a constrained optimization problem: it finds the minimal update modification that acts trivially on the activation subspace. Thus, it serves as a principled generalization beyond the usual synchronous--asynchronous dichotomy: subspace-level protection preserves gradient validity without synchronization stalls, unconstrained-asynchrony instability, or memory overhead that works against pipeline scaling. We demonstrate that FRESCO consistently outperforms representative asynchronous baselines in LLM pre-training, with the gains most pronounced under deep pipelines and high inconsistency levels. Moving beyond the current limitations, FRESCO establishes a new foundation for high-utilization pipeline parallel training, providing a scalable path for large-model development on diverse, resource-constrained systems.

Action-conditioned surgical video generation is a critical yet highly challenging problem for robotic surgery. The core difficulty is that low-dimensional control vectors must precisely govern complex image-space evolution. In this work, we propose a kinematic-to-visual lifting paradigm that converts articulated kinematics into a unified set of five image-aligned control modalities. Building on this representation, we introduce a hierarchically routed visual control framework that selectively activates the most relevant control modalities and motion scales. Instead of uniformly applying all control signals, our model performs hierarchical routing to dynamically allocate conditioning capacity. We further design kinematic-prior-guided routing loss functions to ensure physically meaningful, temporally stable, and efficient expert utilization. To improve efficiency, we propose a budgeted training and inference scheme that leverages routing-induced sparsity. By selectively discarding low-significance control pathways during training and execution, our approach enables adaptive computation that is complementary to standard distillation. We additionally construct a new benchmark with curated articulated annotations, obtained through human-in-the-loop semantic labeling and differentiable pose tracking, providing realistic supervision for action-conditioned surgical video generation. Extensive experiments demonstrate that our method consistently improves action faithfulness, visual fidelity, and cross-domain generalization over diverse baselines. Moreover, our efficient variant achieves substantial reductions in latency while maintaining strong control accuracy.

Existing approaches to optimization modeling using large language models (LLMs) treat tasks through from-scratch construction, generating variables, objectives, and constraints without leveraging reusable structural knowledge. However, in real-world scenarios, expert modelers typically begin by identifying the canonical type or core structure of a problem, followed by iterative refinements based on specific task requirements. To incorporate this expert incremental logic, we propose a novel two-stage framework. In the first stage, we conduct Logic-Anchored Data Synthesis starting from 17 canonical prototypes, preserving critical bottleneck constraints and generating synthetic problem–formulation pairs. These pairs, together with the OR-Instruct dataset, are used for supervised fine-tuning (SFT) to initialize a policy. Separately, for each generated problem, we produce multiple modeling trajectories that follow a prototype-grounded order with variables preceding expressions. These trajectories are ranked according to criteria of accuracy and efficiency, and subsequently used to train a Logical Reward Model (LRM). In the second stage, we introduce Logic-Test-Time Group Relative Policy Optimization (Logic-TGRPO), which leverages the LRM during test-time reinforcement learning to reward accurate prototype identification and disciplined structural adherence while penalizing illogical patterns. Evaluated across seven benchmarks, our 8B parameter model achieves an average accuracy of 79.9%, outperforming comparably scaled baselines and competing effectively with heavyweight multi-agent systems with far fewer parameters. Strong performance on out-of-distribution tasks confirms that the model exhibits genuine incremental reasoning rather than mere memorization. Ablation studies further validate that both stages in our framework are essential.

Prior work has shown that grokked Transformers implement discrete Fourier transforms or group character representations for modular arithmetic. We unify these findings under the Wedderburn--Artin decomposition: using interchange intervention accuracy (IIA), a causal method stronger than ablation, we show that grokked models decompose computation along the Wedderburn components of the target algebra. Across 11 commutative algebras over $\F_p$ ($p = 2,3,5,7$)---including 6 with nontrivial Jacobson radical---all grokked models achieve raw IIA $\geq 0.97$. When models predict incorrectly, errors remain compartmentalized by component ($3.8\times$ chance), confirming that independence is structural, not a byproduct of accuracy. Training dynamics reveal that Wedderburn alignment emerges synchronously with grokking. For non-commutative groups, models exhibit \emph{selective} Wedderburn alignment; non-commutativity raises a capacity threshold that larger models overcome.

Large Language Model-based Automated Algorithm Design (LLM-AAD) has shown promising results across diverse domains by coupling LLMs with iterative search. Some recent efforts have begun to fine-tune LLMs for algorithm design through reinforcement learning. However, these attempts still rely on outcome-level supervision, leaving intermediate ideas and plans without direct credit, even though they determine the conceptual direction of algorithm design. In this paper, we propose Algorithm Tree Policy Optimization (ATPO), which utilizes tree-structured rollouts to decouple the generation of conceptual ideas from their specific code implementations. By applying group-relative advantage estimation, ATPO assigns credit to both strategic ideas and algorithmic implementations, thereby optimizing the policy across multiple levels of abstraction. Integration with FunSearch, EoH, and OpenEvolve demonstrates improvements over their original counterparts. Notably, the learned policy transfers effectively to unseen problem instances and alternative search methods, indicating that it can improve the generalization of algorithm design policies.

Outcome homogenization---models producing similar outputs---has drawn increasing attention in the large language model (LLM) community as researchers and practitioners have come to recognize that alignment practices can reduce model diversity and reproduce dominant narratives. However, most approaches address outcome homogenization exclusively through a technical lens. In this position paper, we argue that outcome homogenization is best understood as a value-chain phenomenon shaped by stakeholders, governance structures, resource allocation, model development, and organizational practices. We identify key drivers of homogenization in the AI ecosystem, showing how current practices reinforce dominant paradigms while marginalizing alternative perspectives. We argue that addressing homogenization requires shifting attention from model-level decisions to the broader power dynamics and organizational practices. Through a case study of LLM vendors’ safety alignment practices, we illustrate how these ecosystem-level forces shape model development and deployment. We conclude with a call to action, offering targeted recommendations for the research community.

Curvature adaptivity is a classical theme in online optimization: for convex Lipschitz losses, adaptive methods interpolate between the optimal $O(\sqrt{T})$ regret for general convex losses and $O(\log T)$ regret under strong convexity. Recent work has shown that Follow-the-Perturbed-Leader (FTPL) achieves optimal $O(\sqrt{T})$ regret even for online non-convex Lipschitz losses, assuming access to an approximate offline-optimization oracle, but these guarantees do not exploit curvature. We show that FTPL can be made curvature-adaptive in the non-convex setting, without knowing in advance how curvature will accumulate over time. Our algorithm replaces the fixed perturbation scale of standard FTPL with a time-varying scale chosen using only past information. We give a simple follow-the-leader tuning rule for this scale and show that it competes, up to constants, with the best choice in hindsight. The resulting method achieves $O(\sqrt{T})$ regret for arbitrary non-convex Lipschitz losses and improves as cumulative curvature grows; with sufficiently accurate oracle calls, it achieves $O(\log T)$ regret when cumulative curvature grows linearly, which includes the classical strongly convex regime. We complement these upper bounds with matching lower bounds for prescribed cumulative-curvature sequences, already for one-dimensional convex losses, showing that the tradeoff between worst-case non-convex regret and curvature-driven fast rates is intrinsic.

Language agents increasingly improve by reusing \emph{skills}---structured procedural artifacts distilled from past experience. In particular, \emph{domain-level} and \emph{model-generated} skills are especially promising. They offer fast adaptation within a domain by encoding domain-specific recurring procedures, and they scale beyond labor-intensive hand-crafting. However, while extraction methods continue to proliferate, understanding remains limited, with no comprehensive study spanning the full skill lifecycle---\textbf{experience generation}, \textbf{skill extraction}, and \textbf{skill consumption}---to ask whether such skills actually work, when they work, and what makes them succeed or fail. To close this gap, we build a utility-grounded evaluation framework that provides systematic experimental results across extractors and target agents, covering five diverse agentic task domains. We find that model-generated skills are beneficial on average but exhibit non-trivial negative transfer, and that neither extractors nor targets behave uniformly. A model can be a strong extractor yet a weak consumer, or vice versa, with skill utility independent of model scale or baseline task strength. To explain these patterns, we then dissect each lifecycle stage in depth, analyzing how experience composition shapes skill quality, what properties characterize useful skills, and how the same skill transfers across different consumers. Finally, we translate these findings into a concrete \emph{meta-skill} that guides skill extraction toward the features tied to actual utility, which consistently improves skill quality across domains and substantially reduces negative transfer.

Multi-step LLM reasoning over structured tables fails because planning and execution share no explicit cell-grounding contract. Existing methods constrain the planner to a left-to-right factorization at odds with table permutation invariance, and score intermediate states by generated content alone, overlooking cell grounding. We conduct a pilot study showing that diffusion language models (DLMs) produce more human-aligned and permutation-stable cell attention on tables than autoregressive models, with a 40.2% median reduction in attention-AUROC variability under row reordering. Motivated by this, we propose TABALIGN, a planned table reasoning framework that operationalizes the contract. TABALIGN pairs a masked DLM planner, whose bidirectional denoising emits plan steps as binary cell masks, with TABATTN, a lightweight verifier trained on 1,600 human-verified attention standards to score each step by its attention overlap with the plan-designated mask. Across eight benchmarks covering table question answering and fact verification, TABALIGN improves average accuracy by 15.76 percentage points over the strongest open-source baseline at comparable 8B-class scale, with a matched-backbone ablation attributing 2.87 percentage points of this gain to the DLM planner over an AR planner on a fixed reasoner. Cleaner DLM plans also accelerate downstream reasoning execution by 44.64%.

Harmful-content detectors are increasingly deployed in safety-critical settings, but they remain vulnerable to subtle rhetorical reframings that preserve the underlying claim while altering its presentation. Existing adversarial robustness methods are typically token-level or prompt-driven, hindering mechanism-level attribution and targeted repair. We propose Adaptive Rhetorical Adversarial Optimization (ARAO), which operationalizes rhetorical theory as an axis-aligned strategy space with directional and interpretable controls for adversarial rewriting. A planner-rewriter pipeline converts strategy vectors into non-contradictory rewrites, enabling mechanism-level failure attribution and axis-targeted repair. Experiments on misinformation and extremist-rhetoric benchmarks show that ARAO improves robustness by 2.35 ROC-AUC points on average over the strongest baselines on attacked sets, while achieving the best performance on original content.

Vision-language models (VLMs) have achieved remarkable progress in 2D visual understanding tasks, yet the passive view-centric perception paradigm limits their capacity for 3D spatial reasoning. Such reasoning often requires actively acquiring spatial evidence beyond the currently visible views, such as camera poses and cross-view spatial relationships, to infer 3D relationships that are not explicitly represented in 2D images. However, existing methods either lack active exploration or merely provide additional views during inference, without explicitly acquiring such spatial evidence beyond 2D images. In this paper, we propose View-to-World (V2W), an active exploration framework that enables VLMs to actively explore 3D worlds and lifts VLMs from passive view-centric perception to active world-centric spatial reasoning. Given a spatial reasoning task with multi-view images, V2W first reconstructs an explicit 3D world with geometry, camera poses, and language-grounded semantics, and exposes it to VLMs through visual and linguistic interfaces. Through multi-turn interaction with these interfaces, VLMs iteratively acquire spatial evidence from the constructed 3D world, enabling active world-centric spatial reasoning. V2W can be integrated with existing VLMs in a training-free manner and can be further enhanced by optimizing the exploration policy via reinforcement learning. Experiments on spatial mental modeling benchmarks demonstrate that V2W substantially improves VLM baselines in the training-free setting and achieves state-of-the-art performance with the learned exploration policy, highlighting the importance of active exploration for spatial intelligence. The code will be available upon acceptance.

We present FRUC, a feed-forward 3D Gaussian splatting framework for dynamic scene reconstruction from uncalibrated collaborative driving views. Existing multi-agent reconstruction frameworks are often hindered by rigid prerequisites, demanding precise spatial calibration and slow per-scene optimization. In this paper, we rethink this task by conceptualizing a distributed multi-vehicle network as a spatio-temporally unstructured ego-centric multi-camera system, where the core challenge lies in enhancing ego-centric occluded geometry through collaboration without degrading the ego's accurately observed visible geometry, while preserving reconstruction efficiency. For efficient reconstruction, FRUC is built upon a visual grounded geometric Transformer backbone to enable one-shot, calibration-free inference from a flexible number of multi-vehicle views. To achieve non-destructive geometric supplementation under uncalibrated cross-agent misalignment, FRUC first introduces an ego-centric causal occlusion field that explicitly derives occlusion evolution as latent priors by modeling agent-wise spatio-temporal correlations. Guided by these occlusion priors, it further formulates cross-agent integration as a deterministic residual denoising process via zero-initialized injection, turning challenging cross-agent fusion into bounded residual learning for robust collaborative blind-spot completion. Through extensive evaluations on the real-world V2XReal and UrbanIng-V2X datasets, FRUC is shown to be a new state-of-the-art for the scene reconstruction of dynamic collaborative driving environments, significantly outperforming existing methods in both rendering quality and efficiency.

Cyclic peptides are an increasingly important therapeutic modality, offering antibody-like binding specificity within compact and chemically tunable scaffolds. Recent deep generative models have advanced target-specific peptide design, where SE(3)-equivariant diffusion and flow-matching frameworks provide natural inductive biases for peptide-protein complex generation. However, existing SE(3)-based peptide generators mainly focus on linear peptides and do not explicitly control cyclic topology. As shown by recent cyclic peptide benchmarks such as CPSea, current generative baselines still have limited cyclization success and remain weak at controllably generating disulfide- and isopeptide-cyclized peptide distributions. In this work, we propose CPFlow, a test-time scaled autoregressive flow-matching framework for receptor-conditioned cyclic peptide sequence-structure co-design. CPFlow autoregressively unmasks residues and predicts full-atom sequence-structure variables with continuous flow matching. By decomposing generation into stepwise flow updates, CPFlow allows each step to balance target binding with structural closure. At inference time, sequence-geometry guided unmasking and multi-particle test-time scaling explore high-confidence autoregressive trajectories beyond the trained model without additional training. Cyclization-aware relative position encodings further enable stable control over head-tail, disulfide, and isopeptide cyclization. Experiments on the CPSea benchmark show that CPFlow outperforms strong generative baselines across designability (scRMSD ↓ 0.30 Å), self-consistency (scRMSD ↓ 0.15 Å), and binding energy (mean ΔG ↓ 3.0), with 93.4% binding success and 0.623 diversity. The code for CPFlow is available at https://anonymous.4open.science/r/CPFlow-F226.

Functional optimization problems are typically solved by optimizing the parameters of a fixed representation, such as a neural network, resulting in highly nonconvex losses that complicate both training and theoretical analysis. An interesting alternative is functional gradient descent (FGD) --- that is, gradient descent directly in function space --- which benefits from strong convergence results and admits a clean theory. However, FGD is difficult to implement in practice because functional gradients are infinite-dimensional, and thus cannot be fully computed nor stored in memory. Existing implementations therefore rely on fixed approximations, which introduce approximation error. We propose a new, theoretically-grounded FGD algorithm that adapts the representation of the functional gradients over the course of optimization. By explicitly incorporating this approximation into the analysis, we establish convergence to a stationary point (for smooth losses) and to a global minimizer (under smoothness + a Polyak-Łojasiewicz-type condition) regardless of our approximations. To the best of our knowledge, this is the first implementable FGD method with such guarantees in a general setting. We demonstrate the effectiveness of our method on regression, numerical solution of PDEs, and modern computer vision. Across settings, our method consistently outperforms both FGD with fixed approximations and neural network baselines in efficiency and accuracy.

Implicit neural representations (INRs) have emerged as powerful tools for encoding signals, yet dominant MLP-based designs often suffer from slow convergence, overfitting to noise, and poor extrapolation. We introduce FUTON (Fourier Tensor Network), which models signals as generalized Fourier series whose coefficients are parameterized by a low-rank tensor decomposition. FUTON implicitly expresses signals as weighted combinations of orthonormal, separable basis functions, combining complementary inductive biases: Fourier bases capture smoothness and periodicity, while the low-rank parameterization enforces low-dimensional spectral structure. We provide theoretical guarantees through a universal approximation theorem and derive an inference algorithm with empirically linear complexity in both the spectral resolution and the input dimension. On image and volume representation, FUTON consistently outperforms state-of-the-art MLP-based INRs while training 2–5× faster.

Off-policy reinforcement learning (RL) for large language models bottlenecks at the rollout stage: every parameter update invalidates the key-value (KV) cache of in-flight agent trajectories, forcing either expensive recomputation or staleness that destabilizes training. Recent advances such as M2PO and asynchronous RLHF tolerate token-level staleness by reweighting the policy gradient with second-moment importance corrections, but they implicitly assume that a stale rollout is sampled from a well-defined behavior policy $\pi_{\theta_{\text{old}}}$. We show that this assumption is silently violated by every modern partial-rollout system: once a stale KV cache is reused under updated parameters $\theta_{\text{new}}$, the resulting token distribution is a hybrid $\pi_{\text{hyb}}$ whose attention projections consume stale keys and values while its MLP, output, and gating projections use the new weights. The token-level importance weight $w_t = \pi_{\theta_{\text{new}}}(y_t \mid x_{

Generative AI is designed to support many tasks, many ways of framing those tasks, and seemingly endless variation in task inputs. But the same flexibility that makes these systems powerful also makes them difficult to evaluate reliably. This flexibility exacerbates the numerous, but often underrecognized and consequential, \emph{researcher degrees of freedom} where even seemingly minor experiment design decisions can substantially affect findings. We demonstrate this problem through an analysis of 36 papers on gender and racial bias in generative AI-assisted resume screening. Although these studies address the same broad research question, they reach strikingly divergent conclusions. To explain this divergence, we adapt recent work on integrative experiment design from the social sciences, and decompose the dimensions along which the studies vary. We then run 1,134 counterfactual experiments by combinatorially varying the experimental modules (names, job-resume pairings, and models) across three baseline experimental frameworks. We find the results prove remarkably brittle: a different design choice flips the conclusion of the reproduced experiment around 22\% of the time. For example, seemingly inconsequential design choices, such as the phrasing of a job description, can reverse the observed conclusion. However, this instability is not always purely stochastic. In some settings, we can build a model that predicts the result of unseen counterfactual experiments with a mean held-out predictive $R^2$ of 0.807. This suggests that despite sensitivities to difference in evaluation design, there exists a learnable structure.

Error Feedback (EF) has become a core mechanism for aggressive gradient compression in communication-efficient distributed training, and is now used in geo-distributed frameworks such as DiLoCo. While existing theory mainly explains its optimization convergence, the generalization behavior of EF remains poorly understood, especially under nonlinear biased compression and distributed local updates. We present the first generalization analysis of EF for both single-node EF-SGD and DiLoCo-EF through an optimization-driven on-average model stability framework, without imposing bounded-gradient assumptions. Technically, we develop a buffer-aware Lyapunov argument that tracks the coupled dynamics of the model and error buffer, while co-coercivity absorbs same-sample gradient-difference energy into the optimization trajectory. For convex losses, our bounds show that compression-induced stability terms are transient, decaying as $\mathcal{O}(\sqrt{b_\alpha}\,T^{-3/4}+b_\alpha T^{-1})$, and therefore do not create a non-vanishing generalization floor beyond the dominant $\mathcal{O}(T^{-1/2})$ optimization term. In the DiLoCo setting, worker averaging further suppresses compression perturbations, and a local stepsize $\eta_l=\Theta(1/(H\sqrt{T}))$ controls the drift from $H$ local steps, yielding the leading $\mathcal{O}(1/\sqrt{MHT})$ excess-risk scaling. Experiments on convex benchmarks corroborate the theory, showing that Top-$k$, Random-$k$, and quantization with matched contraction factor $\alpha$ exhibit nearly indistinguishable generalization trajectories.

The classic paradigm of language identification in the limit models learning as a game between an adversary, who reveals strings from an unknown target language, and a learner tasked with identifying that language. The recently introduced framework of language generation in the limit shifted the objective to better reflect modern language modeling, requiring the learner to produce valid, unseen strings from the target language. Related work highlighted a fundamental tension: a broad coverage of the target often comes at the cost of validity. We introduce a new notion of precision and recast this problem as the classic recall--precision trade-off. We analyze generation in the limit under varying constraints on enumeration, novelty, and validity, aimed at reflecting settings closer to those encountered by large language models. A key contribution is our analysis of learners that are not eventually valid: we allow infinitely many mistakes, provided their frequency tends to zero so that precision remains $1$. We show this relaxation can strictly increase recall when the adversary permanently withholds a large portion of the target language. We also study a continuous relaxation of the novelty constraint, requiring only that a fixed fraction of outputs are novel. Taken together, our results move toward a more realistic model of language generation where occasional errors and repetitions are unavoidable, but their rates are controlled.

Pruning reduces the model size and inference cost of large language models (LLMs), yet preserving performance across diverse downstream tasks remains challenging. Existing structured pruning methods commonly rely on local discrepancy measures or teacher-forced predictive criteria, such as Block Influence (BI) and perplexity (PPL). While these methods are effective on classification benchmarks, the resulting pruned models often degrade severely on generative reasoning tasks. We study redundancy in Transformer-based models and observe that pruning-induced generation drift is strongly correlated with the functional importance of modules, suggesting a more faithful criterion for identifying redundant modules. Based on this insight, we propose GDPruner, a calibration-free, generation-drift-guided block pruning method. GDPruner constructs lightweight self-generated probe trajectories, estimates module importance using tail-position generation drift, and removes redundant modules through adaptive search. Extensive experiments show that GDPruner surpasses state-of-the-art structured pruning baselines while better preserving overall performance and complex generative reasoning ability, offering a promising direction for generation-robust LLM deployment.

We propose a new framework for generative modeling based on a discrete-time stochastic control formulation of measure transport. Adapting classic results from control theory, we formulate our problem as a linear program whose dual variables correspond to the \emph{optimal value function} of the control problem, which directly encodes the optimal control policy. Exploiting this LP formulation, we develop an efficient simulation-free primal-dual algorithm for computing approximately optimal value functions and the associated \emph{value-driven transport} (VDT) policies which approximate the true optimal policy. We show that well-trained VDT policies enjoy numerous favorable properties in comparison with other state-of-the-art methods based on flows, diffusions, or Schr\"odinger bridges: they lead to straight transport paths which can be simulated quickly and robustly, and can be enhanced in all the same ways as diffusion and flow-based models (e.g., conditional generation, classifier-free guidance, unpaired image-to-image translation are all easy to incorporate). We establish several foundational links with the aforementioned models and evaluate our methodology in an extensive range of experiments, all of which indicate strong performance.

Recovering an object's 3D structure and camera motion from a few uncalibrated views is a fundamental problem in computer vision. Classical structure from motion methods recover 3D geometry consistent with the input images but cannot recover unseen regions; native generation methods produce plausible 3D objects which are related but not necessarily faithful to the input images. Methods simply concanetating reconstruction and generation allow error to cascade across the two stages, instead of producing a 3D object consistent with both the distribution and the input. We propose GenSfM, which integrates image-based 3D reconstruction with a native 3D diffusion model. In particular, we use a multi-modal diffusion transformer to jointly denoise the shape latent and per-view pose latents under shared self-attention, grounding generated geometry in the observed views while using object-level priors to infer unobserved regions. A compact 3D-native UV-volume tokenization keeps the joint sequence small as views are added, and a pose-aware stage refines geometry and texture by sampling per-voxel image features at the recovered camera poses. On Toys4k and GSO, for $4$ uncalibrated views, GenSfM improves novel-view PSNR by $3.2$--$4.5$~dB and reduces median Chamfer distance by $2.8$--$3.8\times$ over the strongest 3D-generation baseline, while recovering input-view poses over $96\%$ Acc@$5^\circ$ without any external pose-estimation backbone, post-hoc optimization, or known intrinsics.

Long-context DNA models are limited by token-mixing cost and by how compression allocates representational budget across the genome. Existing approaches operate close to base-pair resolution, apply fixed downsampling, or learn content-dependent chunks without an explicit genomic budget, making long-context pretraining expensive and difficult to control. We introduce GeneZip, a region-aware DNA compression framework that combines H-Net-style dynamic routing with a Region-Aware Ratio (RAR) objective and bounded routing. GeneZip uses static gene-structure annotations during compression training to specify region-wise base-pairs-per-token (BPT) targets; at inference time, it compresses raw unseen DNA without annotations. GeneZip provides three main benefits. First, it is effective: GeneZip variants achieve the best validation PPL among encoder-based compressors, with GeneZip-70M already operating at 137.6 BPT, and across four reproducible DNALongBench tasks---contact map prediction, eQTL prediction, enhancer--target gene prediction, and transcription-initiation signal prediction---GeneZip obtains the best average rank among compared sequence models. Second, it is redundancy-aware: a post-hoc RepeatMasker/TRF analysis shows that, without repeat supervision, GeneZip assigns higher local BPT to TE-derived interspersed repeats and tandem repeats, two major classes of repetitive DNA sequence redundancy. Third, it is efficient: by reducing the effective token-mixing length, GeneZip enables longer-context and larger-capacity pretraining, including 128K-context and 636M-parameter variants on a single A100 80GB GPU, and fine-tunes the eQTL task 50.4$\times$ faster than JanusDNA (50 vs. 2520 minutes). These results establish GeneZip as an effective, redundancy-aware, and efficient compression interface for long-context DNA modeling.

Aligning video generation models with human preference depends critically on the quality of the reward model that scores generated videos. Current video reward models, however, source their reward signal from a model's representational rather than generative capacity: they either repurpose vision-language models as pixel-space scorers, or train auxiliary scoring heads on top of frozen generator features. The generator's actual ability to predict velocities, denoise latents, and produce videos plays no role in scoring its own generations. We argue that this overlooks a structural property of offline preference learning. Preference optimization objective approximating an offline preference dataset mathematically targets a strictly defined KL-constrained optimal policy. By isolating the reward term from this analytical solution and mapping the exact generation log-likelihood to its flow-matching surrogate, the empirical reward emerges deterministically as the relative difference in continuous-time velocity-prediction errors. We formalize this equivalence into GenRM-Flow, a framework where a preference-tuned video generator inherently operates as a process-aware reward model directly in the noisy latent space. We demonstrate that this evaluation mechanism is structurally universal: distinct fine-tuning objectives, whether contrastive like DPO and IPO or non-contrastive like RWR, are merely alternative optimization pathways that share a unified empirical reward readout. Furthermore, the extracted process reward seamlessly substitutes external VLM scorers in Flow-GRPO. Operating purely on native latent outputs without auxiliary RL machinery, this unified generative reward drives measurable improvements in downstream video generation.

Ultra-high-resolution (UHR) remote-sensing imagery captures fine-grained spatial details essential for city-scale Earth observation, yet it exposes a fundamental limitation of current multimodal large language models (MLLMs): native-resolution inputs exceed model capacity, while aggressive downsampling erases the very small objects that many questions hinge on. Generation-understanding unified multimodal models offer a promising remedy, since they can produce intermediate visual observations before answering. However, whether such capabilities genuinely improve UHR remote-sensing understanding remains an open question. Existing UHR benchmarks cannot adjudicate it: they are mostly built from mature datasets covering only a few regions, their image scales remain below real city-level satellite scenes, and they do not systematically identify which tasks benefit from intermediate visual generation. We introduce GeoG2U-Bench, a global benchmark designed to evaluate Generation-to-Understanding (G2U) in UHR remote sensing. GeoG2U-Bench spans 300 cities worldwide, with every raw satellite scene reaching $20{,}000 \times 20{,}000$ pixels, matching the scale at which UHR analysis actually occurs. It comprises 20 subtasks across five capability dimensions and roughly 3,000 expert-verified samples, each requiring models to produce auditable intermediate artifacts such as zoom-in crops, annotated local views, cross-region relation maps, and temporal rewind images. Beyond data, GeoG2U-Bench contributes a dual-protocol evaluation: Direct (answer from the raw input) versus Generate-then-Answer (generate an intermediate visual artifact, then answer), which isolates and quantifies the contribution of generation to downstream understanding. Experiments across several model families reveal that generation is not uniformly beneficial: unfaithful or off-task artifacts can actively degrade performance, whereas reliable, task-relevant artifacts yield consistent gains in cross-scale localization, multi-region comparison, temporal reasoning, and spatial layout understanding. These results show that visual generation can support UHR remote-sensing understanding, but only when it provides reliable visual evidence for the task.

Neural multivariate regression underpins a wide range of domains such as control, robotics, and finance, yet the geometry of its learned representations remains poorly characterized. While neural collapse has been shown to benefit generalization in classification, we find that analogous collapse in regression consistently degrades performance. To explain this contrast, we analyze models through the lens of intrinsic dimension. Across control tasks and synthetic datasets, we estimate the intrinsic dimension of last-layer features ($ID_H$) and compare it with that of the regression targets ($ID_Y$). Collapsed models exhibit $ID_H < ID_Y$, leading to over-compression and poor generalization, whereas non-collapsed models typically maintain $ID_H > ID_Y$. For the non-collapsed models, performance with respect to $ID_H$ depends on the data quantity and noise levels. From these observations, we identify two regimes—over-compressed and under-compressed—that determine when expanding or reducing feature dimensionality improves performance. Our results provide new geometric insights into neural regression and suggest practical strategies for enhancing generalization.

Test-time scaling (TTS) improves large language model (LLM) reasoning by spending additional inference-time compute, most commonly through width-heavy repeated sampling and aggregation. However, this strategy scales cost almost linearly with the number of calls and can remain brittle when the sampled trajectories share the same failure mode. We introduce Maze, a test-time scaling framework that shifts computation away from repeated target-model calls and toward LLM-free planning over ordered few-shot exemplars. The central idea is to treat an ordered exemplar sequence as a prompt trajectory embedded in a latent geometric space. Instead of asking the LLM for many independent attempts, Maze first ranks candidate prompt trajectories using cached encoder representations and a path scorer, ${\it i.e.}$, a lightweight learnable metric, then spends one call (or a few calls) only on the most promising trajectories. This design preserves black-box deployability, since the planner is decoupled from the target LLM and can be implemented with a separate frozen encoder. Across various challenging reasoning benchmarks, single-call Maze consistently outperforms strong single-call prompt-optimization baselines, and MultiMaze recovers a large fraction of width-heavy TTS gains with substantially fewer computational cost measured by number of LLM calls and latency. We further present a theoretical perspective showing when prompt-path ranking can recover much of the benefit of best-of-$N$ style scaling: if high-utility prompt trajectories are sufficiently rankable and sufficiently covered by the explored pool, a deeper planner-side search can substitute for a substantial amount of width-heavy sampling.

Neural operators provide fast surrogates for PDEs but their deterministic predictions limit their use in tasks requiring uncertainty quantification (UQ), especially under geometric variability. Existing approaches primarily model uncertainty in network parameters, largely overlooking the geometry-aware representations learned by the operator itself. We propose REEF-GP (Residual on Embedded Features Gaussian Process), a post-hoc UQ framework that fits a GP to the residuals of a frozen neural operator whose internal embeddings define the kernel feature space. Rather than learning a separate feature map, REEF-GP adapts the operator’s intrinsic coordinate-feature representations to construct geometry-aware uncertainties. To ensure stability and scalability on unstructured domains, REEF-GP incorporates spectral-normalized projections, heteroscedastic geometry-aware noise, and efficient subset-based training that avoids restrictive low-rank approximations. Across five PDE benchmarks with varying geometries, REEF-GP preserves predictive accuracy while achieving calibrated uncertainty estimates competitive with deep ensembles but at a fraction of their cost. Our approach remains robust under geometric distribution shift, with uncertainty concentrating in physically meaningful regions (e.g., shock fronts). Our results demonstrate that accurate and scalable post-hoc UQ for neural operators can be achieved directly in their learned feature space, offering a practical alternative to parameter-centric approaches.

Precise localization of recording electrodes is fundamental to systems neuroscience; yet, current methods depend on labor-intensive post-hoc histology, which is incompatible with real-time feedback or chronic implants. We test the hypothesis that the electrical signals being recorded themselves carry sufficient anatomical information to localize their recording site in 3D atlas coordinates, without the need for histology. To probe this, we introduce a self-supervised framework that learns geometry-aware representations of single-channel local field potentials (LFP), leveraging probe channel geometry as a self-supervisory signal. We compare speech-based SSL objectives (e.g., masked prediction) against geometry-aware objectives that constrain the embedding space to mirror physical channel layout along the probe. These representations outperform masked-prediction baselines, and the gains compound at larger data scales. We test downstream analysis with 3D coordinate regression and brain region classification, benchmarking three SSL backbones (Wav2Vec 2.0, Whisper, and Data2Vec) against supervised baselines (AnyNet, ViT, and classical spectral classifiers) across datasets, laboratories, species, and probe technologies. Our method achieves state-of-the-art results in brain region classification and 3D coordinate regression from raw single-channel LFP, outperforming all baselines, affirming our hypothesis.

We propose a geometry-aware perturbation framework that explicitly models the low-rank, anisotropic, and evolving structure of gradient dynamics to improve generalization in heterogeneous federated learning (HFL). Our approach is built upon three key components. First, we represent perturbations within a low-dimensional subspace that captures the dominant directions of gradient trajectories. Second, we generate perturbations through a structure-aware sampling strategy that aligns with the covariance of projected gradients, enabling distribution-aware exploration. Third, we introduce an adaptive mechanism that dynamically adjusts both the magnitude and the structure of perturbations to match the evolving optimization process. Extensive experiments demonstrate that the proposed method consistently improves generalization across diverse HFL benchmarks, achieving up to a $4.49\%$ improvement on the $\textit{Office10}$ dataset with minimal computational overhead.

Neural operators on irregular meshes face a fundamental tension. Spectral positional encodings, the natural choice for capturing geometry, require cubic-complexity eigendecomposition and inadvertently break gauge invariance through numerical solver artifacts; existing efficient approximations sacrifice gauge symmetry by design. Both failure modes break discretization invariance: models fail to transfer across mesh resolutions of the same domain, and similarly across different graphs of related structure in inductive settings. We propose GIST (Gauge-Invariant Spectral Transformer), a scalable neural operator that resolves this tension by restricting attention to pairwise inner products of efficient approximate spectral embeddings. We prove these inner products estimate an exactly gauge-invariant graph kernel at end-to-end $\mathcal{O}(N)$ complexity, and establish a formal connection between gauge invariance and discretization-invariant learning with bounded mismatch error. To our knowledge, GIST is the first scalable graph neural operator with a provable discretization-mismatch bound. Empirically, GIST sets state-of-the-art on the AirfRANS, ShapeNet-Car, DrivAerNet, and DrivAerNet++ mesh benchmarks (up to 750K nodes), and additionally matches strong baselines on standard graph benchmarks (e.g., 99.50\% micro-F1 on PPI).

Recent 3D geometric foundation models, such as VGGT, provide robust feed-forward 3D reconstruction by directly predicting camera poses and 3D scene points from input images. However, their results remain inaccurate, and scaling them to long sequences or large unordered image sets typically requires chunk-wise processing, which can introduce drift and inconsistency. We present Glob3R, a global SfM-style reconstruction built on 3D foundation models. Our key idea is to explicitly optimize feed-forward geometric predictions. To this end, we augment a frozen Pi3X backbone with a lightweight dense matching head that predicts image warps between selected reference frames and neighboring views. These dense warps are converted into sparse but reliable multi-view feature tracks, which provide correspondence constraints for global optimization. We further introduce a keyframe-based sliding-window association strategy that propagates tracks and relative poses across overlapping windows, enabling scalable reconstruction. Finally, we perform global motion averaging and bundle adjustment to refine camera poses, reduce scale inconsistencies, and recover dense scene geometry. Extensive experiments on indoor, outdoor, large-scale driving, and unordered SfM benchmarks demonstrate that Glob3R achieves robust and accurate reconstruction. It consistently improves over feed-forward foundation-model baselines and recent scalable reconstruction methods, while being more robust than classical SfM pipelines. The refined poses also lead to higher-quality neural rendering, validating the benefit of combining foundation-model priors with global geometric optimization.

We study the global convergence of policy gradient for infinite-horizon entropy-regularized Markov decision processes (MDPs) with continuous state and action spaces. We consider log-linear softmax policies with linear function approximation, which extend the tabular softmax parameterization while retaining a tractable policy class. Under $Q^\pi_\tau$-realizability for the regularized state-action value function, we first establish a non-uniform Polyak--Łojasiewicz (PŁ) inequality. The non-uniformity arises through degeneracy of constants associated with the policy geometry, namely the Fisher information matrix or an uncentered feature covariance matrix. We then identify two feature regimes under which this non-uniform constant can be bounded along the gradient flow. For full-affine-span features, we prove radial unboundedness of the KL regularizer and show that the smallest eigenvalue of the Fisher information matrix remains bounded below by an initialization-dependent positive constant. For simplex-valued features, we prove an analogous radial unboundedness result in the subspace orthogonal to the all-ones vector and obtain a uniform lower bound for the smallest eigenvalue of the uncentered covariance matrix. These results imply global linear convergence of the regularized objective along the gradient flow, i.e. suboptimality decaying as $\mathcal{O}(e^{-Ct})$ for some $C>0$. Our analysis extends the global convergence theory of entropy-regularized softmax policy gradient beyond the tabular setting of Agarwal et al. [2020], Bhandari and Russo [2024], Mei et al. [2020].

Rendering accurate text remains difficult for image generation and editing models, especially when the target contains long, complex, and densely arranged text or rare characters. Existing approaches either improve native text rendering through stronger backbones and data-centric training without explicit glyph priors, or incorporate glyph priors through specialized designs that remain insufficiently accurate and robust under challenging scenarios. We introduce GlyphAnchor, a novel text-rendering enhancement method for both text-to-image and image-editing diffusion transformer models. GlyphAnchor enhances the backbone with lightweight glyph patch conditions whose positions are anchored to the target image through the model's native positional encoding. We train this capability with staged supervised finetuning and further refine it with text-aware post-training to improve robustness. We also introduce InfoTextBench, a benchmark for evaluating text-rich visual text rendering in both generation and editing settings. Experiments across multiple backbones and benchmarks, including long, complex, and densely arranged text and rare character scenarios, show that GlyphAnchor consistently improves text fidelity while preserving overall image quality.

Real-world e-commerce image editing often requires multiple, localized, and auditable operations rather than global restyling. This compositional nature poses a dual challenge: models must precisely apply all requested edits to the correct regions while preserving unmodified content, even under ambiguous instructions. Existing one-shot editors conflate intent resolution, spatial grounding, and synthesis into a single step, frequently resulting in partial execution failures, which is unacceptable for commercial scenarios. To address this, we introduce GMO-E²DIT, an agentic editing framework that couples a Vision-Language Model (VLM) with a mask-conditioned image editor to tackle structured multi-turn task completion. Given an underspecified instruction, the VLM agent constructs a region-grounded edit agenda, effectively decoupling cognitive reasoning from generative rendering. The framework then executes sub-programs via operation-aware masks and references, utilizing a reflection-driven loop to inspect intermediate results and determine the subsequent state. This iterative mechanism reliably preserves safe partial progress, retries unfinished operations, and recovers from errors. Furthermore, we develop a unified data pipeline providing aligned supervision for planning, execution, and reflection, alongside EComEditBench, a comprehensive benchmark for instruction-driven evaluation. Extensive experiments demonstrate that GMO-E²DIT achieves competitive performance compared to strong closed-source models, yielding superior instruction accuracy and edit fidelity over existing baselines.

Moving Object Segmentation (MOS) aims to discover, segment, and track objects that move independently of the camera. Current MOS methods, however, exhibit two fundamental limitations: they rely on pre-computed 2D auxiliary modalities such as optical flow or point trajectories that lack 3D geometric information, and they treat motion as a sequence-level attribute, overlooking the instantaneous motion state of each object. We address both by grounding MOS in 3D space and time, and propose GMOS, a framework that operates directly on RGB video to produce 3D-aware, temporally fine-grained segmentation of multiple moving objects, alongside a foreground--background variant GMOS-S for faster deployment. To support training and evaluation in this regime, we curate GMOS-2K, a dataset of $2{,}210$ real-world videos with per-object temporal motion annotations drawn from five established Video Object Segmentation (VOS) benchmarks, and formalise MOS-I ("I" for *instantaneous*), a temporally fine-grained evaluation protocol with three complementary metrics. GMOS achieves state-of-the-art results across MOS, MOS-I, and Unsupervised VOS benchmarks, while running significantly faster than prior multi-object MOS methods and supporting online inference for streaming deployment.

Multi-objective learning endeavors to concurrently optimize multiple objectives using a single model, aiming to achieve high and balanced performance across diverse objectives. However, this often entails a complex optimization problem to balance the learning of potentially conflicting objectives, leading to solutions with higher memory requirements and computational complexity. This paper introduces a Multi-Objective Goal-Conditioned Supervised Learning (MOGCSL) framework for automatically learning to achieve multiple objectives from offline sequential data. MOGCSL extends the conventional GCSL method to multi-objective scenarios by redefining goals from one-dimensional scalars to multi-dimensional vectors. It benefits from naturally eliminating the need for complex architectures and optimization constraints. Moreover, MOGCSL inherently disentangles uninformative or noisy training instances that fail to achieve desirable long-term rewards across multiple objectives. We also introduce a novel goal-selection algorithm for MOGCSL to model and identify desired and achievable goals for inference. In this paper, we focus on its application to the next action prediction problem in commercial-grade recommender systems. In this context, any viable solution needs to be reasonably scalable and also be robust to large amounts of noisy data that is characteristic of this application space. We show that MOGCSL performs admirably on both counts by extensive experiments. Also, analysis and experiments are included to explain its strength in discounting the noisier portions of training data in recommender systems with multiple objectives. Code is available at: https://anonymous.4open.science/r/MOGCSL-D7A2.

Designing RNA sequences that fold into specific 3D structures is a fundamental challenge in therapeutics. Current RNA inverse folding methods typically condition on 3D backbone coordinates but the generation is limited to 1D discrete sequences. This ignores the continuous side-chain orientations, which are essential for overall structural stability. To address this issue, we explicitly introduce side-chain geometry into the RNA generative process. Unlike co-design frameworks that diffuse across separate raw spaces, we propose Geometric Optimized Latent Diffusion (\textbf{GOLD}), which project sequence and structure into a unified latent manifold. GOLD employs a novel GeoVAE to elegantly integrate discrete RNA sequences with continuous side-chain angles. By mapping these modalities into a single, smooth latent manifold, GOLD naturally facilitates a synergistic co-diffusion process. To enhance structural validity, we further introduce a training-free, gradient-guided sampling mechanism. By decoding the generated sequence probabilities and side-chain geometries into structures during the sampling process, we compute analytical gradients from physical constraints. This dynamic feedback actively guides the generated sequence to a physically stable state. Extensive experiments validate the superiority of GOLD, achieving state-of-the-art performance across sequence recovery, secondary structure validity, and 3D foldability metrics.

The Game of Twenty Questions is a classical problem where a questioner must identify a hidden item by asking binary questions. Although simple in form, the game is strategically challenging, especially when the questions are expressed in natural language. It is also a useful abstraction for realistic information-seeking tasks, such as medical diagnosis and troubleshooting. Existing approaches often rely on simplifying assumptions that degrade worst-case performance. This is an issue with serious implications in high-stakes applications. In this work, we introduce Strategic Language Search (SLS), an adversarial counterpart of Twenty Questions, and formalize it and its variants as two-player zero-sum extensive-form games. We then propose Game of Thought (GoT), a framework that applies game-theoretic techniques to approximate a Nash equilibrium (NE) strategy for the restricted variant of the game. Empirical results demonstrate that our approach consistently improves worst-case performance compared to (1) direct prompting-based methods and (2) heuristic-guided search methods across all tested settings.

Transformers excel at sequence modeling, yet Softmax attention incurs quadratic complexity and unbounded KV cache growth. While linear attention offers a promising alternative, existing approaches lack systematic functional comparison with Softmax attention, rigorous error analysis, and a theoretically grounded improvement roadmap. We address this gap by framing linearization as KV cache approximation and establishing a principled pathway from Softmax attention to linear models. Our analysis identifies five critical components—redundancy elimination, token-level quantization with positional separation, positional compression, inter-layer similarity, and multi-state decomposition—each accompanied by theoretical justification and error bounds, with explicit connections to existing mechanisms. Building upon this framework, we introduce GPA, a linearized attention model that inherits pretrained weights and achieves state-of-the-art results. GPA outperforms strong baselines including MVA and GSA across multiple benchmarks, while requiring less fine-tuning resources. Our work provides both theoretical clarity and practical guidance for advancing linear attention, charting a principled course toward efficient, scalable alternatives to Softmax attention.

Oracle-guided biological sequence design must improve predicted function without moving outside the sequence distribution where the oracle is trustworthy. We introduce Generative Population Annealing GPA, a test-time sampler for sequence design with a frozen generator and frozen oracle. GPA instantiates annealed Sequential Monte Carlo at population scale: particles are initialized from a pretrained sequence prior, reweighted by an oracle reward tilt, selectively upsampled as effective sample size falls, mutated with the pretrained generator, and returned as the design pool. The base sampler targets a reward-tilted prior; practical variants shape the proposal or modify the objective to trade activity, specificity, fidelity, and diversity. Across enhancer and promoter benchmarks, GPA scales to thousands of sequences and is competitive with inference-time samplers, gradient-based editors, tree-search diffusion, and RL-fine-tuned generators. The same inference loop is evaluated with masked discrete-diffusion and autoregressive DNA generators. GPA reaches high predicted activity while preserving strong motif fidelity and model-based likelihood, although specialized baselines retain higher diversity or stronger $k$-mer fidelity in some settings. Cross-oracle and sequence-level audits suggest fewer obvious reward-hacking pathologies, including shorter homopolymer runs than CTRL-DNA in the HepG2 audit (5.1 bp mean maximum run versus 20.9 bp), but all validation remains computational.

We consider the well-studied setting of minimizing a convex Lipschitz function using either gradient descent (GD) or its stochastic variant (SGD), and examine the last iterate convergence. By now, it is known that standard stepsize choices lead to a last iterate convergence rate of $\log T/\sqrt{T}$ after $T$ steps. A breakthrough result of Jain et al. [2019] recovered the optimal $1/\sqrt{T}$ rate by constructing a non-standard stepsize sequence. However, this sequence requires choosing $T$ in advance, as opposed to common stepsize schedules which apply for any time horizon. Moreover, Jain et al. conjectured that without prior knowledge of $T$, no stepsize sequence can ensure the optimal error for SGD's last iterate, a claim which so far remained unproven. We prove this conjecture, and in fact show that even in the noiseless case of GD, it is impossible to avoid an excess poly-log factor in $T$ when considering an anytime last iterate guarantee. Our proof further suggests that such (slightly) suboptimal stopping times are unavoidably common.

Pretrained model weights are increasingly released under commercial licenses, usage restrictions, or other conditions that prohibit unauthorized fine-tuning. In practice, however, such misuse is difficult to detect or verify after the fact. This motivates a stronger objective for protected weight release: weights should remain useful for intended inference, yet become practically unattractive to repurpose through unauthorized gradient-based adaptation. We frame this objective as a \emph{scorched-earth} strategy in parameter space: rather than only proving infringement after misuse occurs, the released weights themselves should react destructively when unauthorized fine-tuning begins. To realize this idea, we propose \textbf{Gradient-Mine Units (GMUs)}, a data-free weight-space protection mechanism for pretrained networks. GMUs are planted into selected feedforward layers as hidden units with extreme internal scale, while a locking mechanism keeps them silent at initialization so that the original inference behavior is preserved. During fine-tuning, this locked state is progressively broken, allowing the planted units to emit amplified gradients that disrupt the model's native adaptation dynamics. We formulate GMUs in a general feedforward setting and show that gated instantiations naturally provide an additional hard lock. Empirically, we validate the method on both large language models and Vision Transformers. Across multiple architectures and downstream tasks, GMUs preserve pre-fine-tuning utility while substantially degrading or destabilizing standard fine-tuning. These results suggest that protected weight release can move beyond post hoc attribution toward a practical deterrence mechanism for unauthorized adaptation. Our implementation is available at here.

A major recent advance in quantization is given by microscaled 4-bit formats such as NVFP4 and MXFP4, quantizing values into small groups sharing a scale, assuming a fixed floating-point grid. In this paper, we study the following natural extension: assume that, for each group of values, we are free to select the "better" among two or more 4-bit grids marked by one or more bits in the scale value. We formalize the power-of-two-grids (PO2) problem, and provide theoretical results showing that practical small-group formats such as MXFP or NVFP can benefit significantly from PO2 grids, while the advantage vanishes for very large groups. On the practical side, we instantiate several grid families, including 1) PO2(NF4), which pairs the standard NF4 normal grid with a learned grid, 2) MPO2, a grid pair that is fully learned over real weights and activations, and 3) SFP4, a TensorCore-implementable triple which pairs NVFP4 with two shifted variants. Results for post-training quantization of standard open models and pre-training of Llama-like models show that adaptive grids consistently improve accuracy vs single-grid FP4 under both weight-only and weight+activation quantization.

Long-video understanding in VLMs is bottlenecked by a single monolithic forward pass over thousands of frames at quadratic attention cost. A common mitigation is to first $select$ a small subset of informative frames before the forward pass; common for training-free selectors via auxiliary encoder-space similarities. Such signals are capped by contrastive pretraining, which usually fails on reasoning-heavy queries (negation, cross-frame counting, holistic summarization). We propose $\textbf{GridProbe}$, an efficient training-free posterior-probing inference paradigm that scores evidence in $\textbf{answer space}$ using a frozen VLM's own reasoning and then selects question-relevant frames $\textbf{adaptively}$, resulting in sub-quadratic attention cost with little to no accuracy loss. We arrange frames on a $K{\times}K$ grid and run lightweight row $R$ and column $C$ probes, where each probe reads its peak posterior as a query-conditioned confidence. The outer product of $R$ and $C$ yields an interpretable $\textbf{importance map}$ whose skewness and kurtosis drive $\textbf{Shape-Adaptive Selection}$, a closed-form rule that reliably replaces the fixed frame budget $M$ with a per-question $M_{\mathrm{eff}}$. We show empirically that $M_{\mathrm{eff}}$, surprisingly, tracks intrinsic question difficulty without ever seeing the answer, a sign of test-time adaptive compute. On Video-MME-v2, $\textbf{GridProbe}$ matches the monolithic baseline within $1.6$ pp Avg Acc at $3.36\times$ TFLOPs reduction, while on LongVideoBench it Pareto-dominates the baseline ($+0.9$ pp at $0.35\times$ compute). Because the selector and QA models can be decoupled, pairing a small 2B selector with a stronger 4B or 8B QA is strictly Pareto-dominant over the 2B monolithic baseline (up to $+4.0$ pp at $0.52\times$ compute, on average), with no retraining. Finally, the interpretability of the importance maps opens future avenues for behavioral diagnostics, grounding, and frame-selection distillation.

Multi-Agent Debate (MAD) improves the reasoning performance of Large Language Models (LLMs) through multi-round interaction. However, LLMs in MAD are highly susceptible to blind conformity. Existing individual evaluation methods, typically based on confidence or perplexity, fail to reflect the correctness of reasoning and may even exacerbate blind conformity. To address this, we shift the perspective from individual evaluation to group interaction. We define mutual referencing among LLMs as \textbf{Debate Relationships} and recognize that regulating these relationships is the key to mitigating blind conformity. In this paper, we propose a novel framework for \textbf{D}ynamically r\textbf{E}gulating deb\textbf{A}te \textbf{R}elationships (DEAR) from the group perspective. At first, DEAR quantifies consensus and divergence as \textit{group evidence} to capture the debate state. Then, DEAR operates through three stages: 1) What: perceiving group consultation tendency and uncertainty; 2) Who: introducing a Selection RL-Agent to dynamically select reference peers; and 3) How: adopting a Behavior RL-Agent to adaptively adjust generation behaviors.

Spatial and temporal resource constraints are critical for both biological and artificial intelligent systems. Here we define differentiable cost terms for breadth, depth, and time within a recurrent convolutional neural network conceived as a finite subset of an infinite lattice architecture. We optimize these costs jointly with task errors via backpropagation, and efficient computational graphs emerge organically through training. We find that breadth, depth, and time can be traded off against each other to achieve a given level of performance. Networks grow in all three dimensions with task complexity and spontaneously take more recurrent steps when inputs are occluded. Surprisingly, time used by the model correlates with human reaction times in an object recognition task. Our framework provides a normative account of how resource constraints shape neural architectures, connecting to questions about brain design in neuroscience, and may help illuminate the diversity of neural solutions found in nature.

There is a growing need for understanding how trained machine learning models behave beyond standard predictive performance. With this work, we aim to understand trained machine learning models by questioning their data preferences. We propose a mathematical framework for guided data generation that allows us to produce fixed-label, prediction-risky, parameter-sensitive, or model-contrastive samples, among others. To showcase our framework, we pose these queries to a range of models trained on a range of classification and regression tasks, with answers in the form of generated data.

Robotic manipulation with dexterous hands is a cornerstone of Embodied AI, yet its progress is stifled by the high cost of collecting embodiment-aware teleoperation data. While abundant egocentric videos of human hands offer a scalable alternative, the profound discrepancies in appearance, articulation, and camera viewpoints between human and robotic data raise significant challenges for co-training. Though existing general image-editing models demonstrate strong capabilities, they lack necessary embodiment-specific priors to fully bridge this gap. In this work, we present HandEdit, a unified large-scale embodiment-aware image-editing dataset and benchmark specifically designed to transform human hands and arms into various dexterous robotic embodiments within egocentric frames. HandEdit comprises over 200M editing instances derived from five diverse source datasets, covering 26 distinct URDFs, including 13 hand-only and 13 hand-arm configurations. Alongside the dataset, we establish a unified benchmark protocol with two tracks: Hand-only and Hand-Arm, supporting URDF-conditioned evaluation. We conduct extensive evaluations of 11 representative image-editing baselines using a multi-dimensional metric suite, including generic similarity metrics, VLM-based judgment, and embodied-aware metrics. HandEdit serves as a critical resource at the intersection of image editing and robotics: it advances embodiment-aware editing models while enabling scalable dexterous robotic learning from abundant human video data, paving the way for more generalizable Embodied AI. Code, datasets, and benchmarks are available at \url{https://anonymous.4open.science/r/HandEdit}.

The rapid growth of large language models is driving organizations to expand their GPU clusters, often with GPUs from multiple vendors. However, current deep learning frameworks lack support for communication across heterogeneous vendor GPUs, leading to inefficiency and higher costs. We present HetCCL, a collective communication library that unifies vendor-specific backends and enables RDMA-based communication across GPUs without requiring driver modifications. HetCCL introduces two novel mechanisms that enable cross-vendor communication while leveraging optimized vendor libraries, NVIDIA NCCL and AMD RCCL. Evaluations on a multi-vendor GPU cluster show that HetCCL matches NCCL and RCCL performance in homogeneous setups while extending support to heterogeneous environments, enabling practical and efficient training across NVIDIA and AMD GPUs without changes to existing deep learning applications.

TrueSkill is a widely used Bayesian rating system for inferring latent player skill from game outcomes. However, its standard formulation assumes a fixed global performance-noise parameter, making it sensitive to atypical outcomes and unable to distinguish skill from player-specific consistency. We propose two heteroscedastic extensions: a match-specific precision model that yields a robust heavy-tailed comparison likelihood, and a player-specific precision model that captures differences in performance consistency among players. Both extensions use Gamma-distributed precision variables and retain a modular factor-graph representation. Because the introduced Gaussian-Gamma factors do not admit the same closed-form expectation-propagation updates as vanilla TrueSkill, we derive an efficient hybrid message-passing scheme that combines expectation-propagation updates for outcome truncation factors with variational updates for latent precision factors. Experiments on synthetic and real match data show that the proposed models improve robustness to anomalous outcomes and provide interpretable estimates of player consistency.

Long-horizon visual search requires a vision-language policy to execute multi-step spatial actions over high-resolution images, where performance depends on the entire trajectory rather than any single decision. Existing supervision breaks at this horizon: outcome-only RLVR collapses many geometric decisions into a single delayed signal and discards the per-step geometry the environment already computes, while LLM-judged step rewards recover density at the cost of verifiability, scale linearly with trajectory length, and conflate distinct failure modes. We propose HSSP, a self-play framework in which a Hider adaptively weights challenging skeletons from a fixed trajectory pool and a Seeker learns to solve them through grounded spatial actions. Instead of relying on external judges, HSSP supervises every step with a geometry-verifiable process reward built from target-directed progress and exploration coverage, with trajectory length and spatial redundancy enforced as separate Constrained Markov Decision Process (CMDP) constraints. We pair HSSP with V-Trace, a trajectory-level companion dataset that augments skeleton seeds from established benchmarks with target-region annotations and explicit search-regime labels, ready for any trajectory-level training framework. The reward design admits closed-form guarantees on backtracking, find-versus-explore separation, and policy invariance, providing theoretical backing for the observed gains. Extensive empirical results on eleven challenging multimodal benchmarks, supported by controlled ablations and qualitative case studies, show that HSSP consistently outperforms advanced baselines and transfers zero-shot to OOD tasks.

Recent think-answer approaches in VLMs, such as Qwen3-VL-Thinking, boost reasoning performance by leveraging intermediate thinking steps before the final answer, but their high computational cost limits real-world deployment. To distill such capabilities into compact think-answer VLMs, a primary objective is to improve the student's ability to utilize visual evidence throughout its reasoning trace. To this end, we introduce a novel think-answer distillation framework that encourages the student to anchor its thinking on visual information by masking the student's salient reasoning prefixes. To compensate for such masked textual cues, the student is encouraged to rely more on visual evidence as an alternative source of information during distillation. Our masking strategies include: 1) $\textit{token-wise salient reasoning-prefix masking}$, which masks high-influence reasoning prefixes selectively for each next-token prediction, and 2) $\textit{self-paced masking budget scheduling}$, which gradually increases the masking scale according to distillation difficulty, {measured by discrepancy between teacher--student distributions. In the distillation phase, the student is guided by our salient reasoning-prefix mask, which blocks both future tokens and salient reasoning cues, in place of the standard causal mask used for auto-regressive language modeling. Experimental results show that our approach outperforms recent open-source VLMs, VLM distillation, and self-distillation methods on multimodal reasoning benchmarks, while further analyses confirm enhanced visual utilization along the student thinking process.

Conditional independence (CI) testing is a fundamental problem in statistics. However, classical nonparametric CI tests suffer from the curse of dimensionality, as nonparametric estimation degrades rapidly in high dimensions. This paper proposes RP-CIT, a random-projection-aggregated CI test that remains effective in the high-dimensional regime. The central idea is to project $X$ and $Y$ onto random one-dimensional subspaces before applying a univariate base CI test. This projection (i) preserves conditional independence under the null hypothesis and (ii) reduces the original high-dimensional problem to a collection of univariate CI subproblems, thereby avoiding high-dimensional nonparametric estimation. Since a single projection may miss the dependence signal, we aggregate the $p$-values from $k$ independent projections via a Bonferroni-max rule, which asymptotically controls the Type~I error at the nominal level and yields a limiting Type~II error bound that decreases exponentially in $k$ under a detectable-projection condition. The paper also develops a top-$r$ aggregation extension for alternatives whose signal is spread across multiple projection directions. More broadly, the random-projection aggregation framework can be combined with other scalar base tests; as one example, we use a projected GCM base test to accommodate higher-dimensional conditioning variables. Experiments on synthetic benchmarks, image-valued CI tasks, and Norman Perturb-seq module skeleton learning illustrate the calibration, power, and practical utility of the framework.

Differentially Private Stochastic Gradient Descent (DP-SGD) is the dominant approach to private deep learning, but it pays compute and privacy budget \emph{per trained model}: each new architecture, ensemble member, or downstream fine-tuning round incurs additional privacy cost. We propose \textsc{SPS} (Summarize--Privatize--Synthesize) and its enhanced variant \textsc{SPS+}, dataset-distillation algorithms that release a private synthetic dataset by privatizing intermediate activation statistics from a public pretrained model. Once released, the dataset is post-processed freely: any downstream architecture, ensemble, federated aggregation, or continual update incurs \emph{zero} additional privacy cost. In particular, a single \textsc{SPS+} dataset distilled from a Wide ResNet-22-8 transfers zero-shot to architectures with different inductive biases (Vision Transformers, Swin Transformers, and ConvNeXt), all without re-incurring privacy. Empirically, \textsc{SPS+} is competitive with state-of-the-art DP-SGD across $\epsilon \in \{1,2,4,8\}$ on CIFAR-10 and CIFAR-100, with a notable advantage on CIFAR-100 under strict privacy, achieving a $5.8\%$ gain at $\epsilon{=}1$ over compute-matched DP-SGD. \textsc{SPS+} additionally outperforms prior generation-based DP methods by a large margin and supports private federated and continual learning out of the box.

Markovian stochasticity naturally arises in many modern learning problems, including optimization with dependent data streams and reinforcement learning. Despite extensive research on convergence of stochastic gradient methods under such stochasticity, high-probability convergence remains largely unexplored. To close this gap, we provide the first high-probability convergence guarantees for gradient-based methods under Markovian stochasticity. We study SGD with a specialized mini-batch-based gradient estimator and establish high-probability convergence results in the non-convex setting under the classical bounded stochasticity assumption, and further extend the analysis to a weaker bounded-variance assumption by incorporating gradient clipping, yielding convergence guarantees for accelerated stochastic gradient descent in the convex setting. For both methods, we characterize the oracle complexity via a concentration inequality that explicitly captures the effect of Markovian dependence on the resulting convergence rates.

Large reasoning models achieve high accuracy through extended chain-of-thought but generate 5--8 more tokens than necessary, applying verbose reasoning uniformly regardless of problem difficulty. We propose Hint Tuning, a data-efficient approach that teaches models to calibrate reasoning depth. Our key insight: the corresponding instruct model serves as an ideal difficulty probe. By testing what the instruct model can solve with varying guidance, we automatically construct training data across three states: No-Hint (direct answer), Sparse-Hint (minimal prefix), and Full-Hint (complete reasoning). This converts the abstract challenge of difficulty labeling into a measurable consistency check between the instruct and reasoning models. With only 1K self-annotated samples, Hint Tuning achieves 24--66% token reduction (31.5% average) across mainstream reasoning models (Qwen3-Thinking, DeepSeek-R1-Distill) at multiple scales (4B--32B) while maintaining competitive accuracy on five benchmarks. Unlike methods requiring massive distillation datasets or expensive RL, we achieve superior efficiency through simple alignment with the instruct model's capabilities.

A central bottleneck in multi-hop Question Answering (QA) is that the granularity at which a question is expressed often differs from the granularity at which corpus evidence is retrievable. Existing methods address this mismatch either by imposing fixed graph structures over the corpus or by iteratively reformulating the query, but these strategies do not explicitly decide when a query unit is already supported by evidence and when it should be refined. We formulate this bottleneck as retrievable granularity discovery and introduce Hi-Q, an evidence-conditioned framework for hierarchical query refinement. At each query node, a resolution operator tests whether retrieved evidence supports the current query unit; resolved nodes terminate, while unresolved nodes are expanded by a dependency-preserving binary operator and checked by a semantic coverage verifier. Hi-Q therefore grows a query tree whose topology is determined by corpus support signals rather than by a fixed decomposition template or a pre-built graph. Across three multi-hop QA benchmarks, Hi-Q achieves 57.9 EM and 69.3 F1 on average, outperforming PropRAG, a graph-based RAG baseline, by 5.6 EM / 3.9 F1 and IRCoT, an iterative retrieval baseline, by 13.7 EM / 15.8 F1. Under full-corpus retrieval, Hi-Q maintains its gains with 53.4 EM and 65.2 F1 on average, improving over IRCoT by 16.3 EM / 19.4 F1 without corpus-wide graph construction.

Deployed approaches for AI text detection often rely on training-time access to labeled datasets of both human-written and AI-generated text. This approach is vulnerable to three types of distribution shifts that occur continually post-deployment, and for which labeled data is often unavailable: adversarial humanization, new LLMs being released, and temporal drift in human writing. Simultaneously, existing approaches do not leverage a key signal of LLM usage: inference-time homogeneity. We propose a test-time adaptation (TTA) approach, using semi-supervised learning, that adapts to distribution shifts by leveraging homogeneity among unlabeled samples observed at inference time. Empirically, we find that state-of-the-art supervised detectors systematically fail when they encounter distribution shifts in AI-generated and human writing, both adversarial and natural, while test-time adaptation with semi-supervised learning is largely robust; e.g., the commercial model Pangram detects just 24.1\% of our adversarial AI-generated text, compared to 90.5\% for our test-time approach. We establish that test-time adaptation is a promising framework for AI text detection in the wild.

Unified Multimodal Models (UMMs) integrate visual generation and understanding within a shared parameter space, yet existing post-training typically relies on external annotations and supervised post-training data, and rarely couples these two capabilities for iterative improvement. We propose Holistic EvoLution via Intrinsic eXchange (HELIX) for Unified Multimodal Models, a two-stage cyclic post-training framework that forms a double-helix coupling between generation and understanding: generation produces controllable visual evidence while understanding provides semantic judgments. A target-driven Data Bridge connects the two stages and supplies reliable training signals for both generation and understanding updates. In Stage 1, the understanding branch provides a self-evaluated likelihood gain reward to optimize the generative policy for stronger semantic faithfulness. In Stage 2, the generation branch supplies reconstruction-based visual evidence, and we refine the understanding branch with Reconstruction and Prompt-Alignment rewards. Extensive experiments on Bagel, trained at 512px, demonstrate consistent gains on compositional and knowledge-intensive benchmarks, outperforming the baseline by +5.6 on T2I-CompBench, +7.1 on GenEval, +1.54 on DPG-Bench and +0.02 on WISE when evaluated at 1024px.

Vision-language models (VLMs) show strong multimodal capabilities but still struggle with fine-grained vision-language reasoning. We find that long chain-of-thought (CoT) reasoning exposes diverse failure modes, including perception, reasoning, knowledge, and hallucination errors, which can compound across intermediate steps. However, most existing data for reinforcement learning with verifiable rewards (RLVR) does not involve complex reasoning chains grounded in visual evidence throughout, leaving these failure modes understressed during training. We therefore propose HopChain, a scalable framework that constructs an out-of-distribution (OOD) proxy task: OOD only in query format, while strengthening foundational abilities shared with downstream benchmarks to support generalizable gains. Concretely, HopChain synthesizes multi-hop vision-language reasoning chains, each a logically dependent chain of instance-grounded hops where earlier hops establish the instances, sets, or conditions needed for later hops, ending in an unambiguous numerical answer for verifiable rewards. In experiments, we train Qwen3.5-35B-A3B and Qwen3.5-397B-A17B under two RLVR settings (with and without HopChain's multi-hop data) and compare them across 24 benchmarks spanning STEM and Puzzle, General VQA, Text Recognition and Document Understanding, and Video Understanding. Although the multi-hop data is not designed for any specific benchmark, it improves 20 of 24 benchmarks on both models, indicating broad and generalizable gains. Consistently, replacing full chained queries with half-multi-hop or single-hop variants reduces the average score across five representative benchmarks from 70.4 to 66.7 and 64.3, respectively. Improvements are particularly substantial in long-CoT and ultra-long-CoT regimes, peaking at more than 50 accuracy points in the ultra-long-CoT regime. These results show that OOD proxy tasks for long-chain visual reasoning are a scalable RLVR supervision source for generalizable VLM reasoning.

Estimating physical pressure from vision is essential for understanding contact-rich hand-object interaction. However, prior vision-based pressure estimation methods are largely limited to planar surfaces and single image input, making them difficult to apply to dynamic hand-object interaction with diverse objects. We instead formulate pressure estimation as a hand-centric video prediction problem with monocular video as input. This formulation predicts temporally evolving per-vertex normal pressure and contact directly on the hand mesh, yielding a unified output space independent of object shape and sensor layout. Building on this formulation, we propose HOPE, a framework with two key components. First, we lift tactile-glove pressure, planar-sensor pressure, and distance-based hand-object contact annotations into a shared hand vertex space, allowing bare-hand contact data to regularize pressure learning where metric labels are unavailable. Second, we introduce a vertex-anchored video transformer that treats each vertex as a persistent token, aggregates visual features and hand pose over time, and uses a contact-gated pressure head to enforce that pressure vanishes without contact. Experiments on OpenTouch, PressureVisionDB, and hand-object contact benchmarks demonstrate that HOPE supports a common hand-centric representation across object-pressure, surface-pressure, and contact-supervised HOI settings. Despite using metric pressure supervision primarily from gloved-hand videos, HOPE generalizes to bare-hand egocentric and in-the-wild videos, producing joint contact and pressure predictions beyond the scope of contact-only or planar-pressure baselines.

For graph node classification, calibrated class probabilities are need when confidence scores, usually the maximum predicted class probability, are used to rank prediction, defer uncertain nodes to human review or control risk. Existing post-hoc calibrators either apply one global temperature or use graph-awared modules without a derived strctural form. We study node-level calibration for graph node-classification problem under structural heterogeneity. Our first results show that a logit-only temperature rule is insufficient when nodes with identical logits but different local homophily require different optimal temperatures. We then analyze a population-concentration contextual stochastic block model with Gaussian features and a one-layer linear GCN. Under equidistant class means, the Bayes posterior over class-template scores is a temperature-scaled softmax whose inverse temperature is governed by homophily-dependent signal strength. In the positive-signal homophilic regime, the resulting temperature decreases approximately with normalized local homophily. This law motivates Homophily-aware Temperature Scaling (HoTS), a simple post-hoc calibrator that assigns each node a positive scalar temperature from entropy-based logit concentration and estimated local homophily. HoTS has three parameters, preserving class, and learn the strength of the structural correction from calibraion data. Across 18 node-classification benchmarks, two GNN backbones, and eight calibration baselines, HoTS achieves the best mean Expected Calibration Error (ECE) of $4.66\%$, the best average rank, and the strongest confidence ranking in selective classification.

Feature learning is widely viewed as the mechanism that distinguishes neural networks from fixed-kernel methods, yet its effect on the underlying function space remains poorly understood. We aim to precisely characterize how the function space (e.g., RKHS) endowed by a two-layer neural network during gradient descent training. We prove that, in a high-dimensional proportional regime, the post-update feature distribution is well approximated by a target-dependent spiked Gaussian covariance, yielding a deformed kernel that can be written as the original isotropic kernel evaluated on geometrically transformed inputs. This characterization enables a spectral analysis of the corresponding integral operator. We prove that the global eigenvalue decay rate is preserved, while the spike selectively alters the leading eigenspaces. For ReLU activations, we derive an explicit perturbative expansion showing that feature learning boosts the eigenvalue of the linear eigenfunction aligned with the target direction and mixes the top radial eigenfunction with a target-aligned quadratic harmonic. These results provide a precise function-space description of early feature learning: gradient descent does not merely rescale a static kernel, but induces a data-adaptive deformation that preferentially enriches directions aligned with the teacher signal.

Message-passing graph neural networks (MPGNNs) are commonly compared with the Weisfeiler-Lehman (WL) color-refinement procedure, but this comparison does not quantify the resource parameters a network needs to realize color refinement with bounded-size messages and finite numerical precision. We study the cost of simulating a single color-refinement step on unattributed graphs. We distinguish input-independent, or oblivious, simulation from instance-dependent simulation. In the former, the parameters, or their distributions in randomized models, are fixed before the input instance is known. Our results show that the local form of WL color refinement hides a global relabeling problem. In the oblivious setting, deterministic and zero-error randomized MPGNNs cannot solve this problem in the worst case using only shallow networks with small messages. We complement this lower bound with a nearly matching construction in a stronger rooted, port-aware model. By contrast, when the color set is large, bounded-error randomness can greatly reduce the cost, and a one-layer MPGNN with messages of logarithmic size and a logarithmic number of random bits suffices. We show that this logarithmic number of random bits is essentially necessary for shallow, small-message simulations. When the color set is small, we still obtain a rooted, port-aware simulation, but this construction requires more layers or larger messages. We also prove that this extra cost is partly unavoidable, as small color sets force a nontrivial trade-off between the number of layers and the message size. Finally, instance-dependent simulation can be much shallower, but the required instance-specific parameters are not necessarily easy to find. Together, these results reveal quantitative structure hidden behind the statement that MPGNNs match WL color refinement.

Multi-task benchmarks have become a central pillar of machine learning research, yet their growing influence has incentivised benchmark gaming -- strategic actions taken to improve the leaderboard rank of a specific model. Treating datasets as voters and models as candidates, we consider benchmark-specific training -- the inclusion of benchmark data in training -- as a form of election manipulation. For any ordinal benchmark, the problem of choosing datasets to train on so that a target model becomes top-ranked corresponds to shift bribery, a class of manipulation problems from computational social choice. Leveraging this identification, we show that the benchmark-specific training problem is NP-hard under Borda count and mean win rate. Complementing this worst-case perspective, we introduce the instance-level robustness, the minimum number of datasets a model developer must include in training to top a given leaderboard, and derive expressions for it under arithmetic mean, median, mean win rate and pairwise majority. We evaluate these expressions on MMLU under HELM and on BIG-Bench Hard (BBH) under the Open LLM Leaderboard. Across both suites, mean win rate is hardest to manipulate: this gap is clear on BBH (24 tasks, 4507 models), where its median robustness is 22 tasks (92\%), compared with 13 (54\%) under arithmetic mean and 12 (50\%) under median and pairwise majority.

We study signal propagation in linear recurrent models at finite width. While existing signal propagation theory relies predominantly on the infinite-width limit, it remains unclear for how long that approximation remains accurate when recurrent depth $t$ grows jointly with width $n$. This question is especially relevant for modern recurrent sequence models, whose natural operating regime involves long input sequences, i.e., large $t$. We derive exact finite-width formulas for the hidden state signal energies in linear recurrences under complex Gaussian initialization. Using these formulas, we identify the joint depth-width scaling regimes that govern signal propagation: (i) a *subcritical regime* $t=o(\sqrt n)$, in which the infinite-width approximation remains valid; (ii) a *critical regime* $t\sim c\sqrt n$, in which non-negligible deviations from infinite-width predictions appear and a nontrivial joint scaling limit emerges; and (iii) a *supercritical regime* $t\gg \sqrt n$, in which finite-width effects dominate. Thus, our results pinpoint the precise recurrent depth scale at which infinite-width theory breaks down in long-range linear recurrences. In turn, this shows when standard initialization schemes, such as Glorot, become unstable. More broadly, our results demonstrate that finite-width effects accumulate more rapidly with depth in recurrent models than in feedforward ones, leading to qualitatively different signal propagation behavior.

Large language models generate one token at a time, yet their responses show remarkably consistent length structure: step-by-step solutions converge in predictable token counts, retrievals stop after a few sentences, retractions extend responses by measurable amounts. We ask whether the model carries an internal estimate of how much response remains. Training minimal-capacity linear probes on frozen hidden states of three open-weight 7-8B models across seven completion-style datasets, we find three converging pieces of evidence. First, total response length is linearly decodable from the prompt's last hidden state alone, before any output is emitted. Second, probe directions trained on natural-language datasets transfer broadly, including to controlled synthetic completions never seen in training, outperforming a statistical baseline; the converse direction generally fails, and this asymmetry is itself informative. Third, on curated high-loss completions, the probe's per-position estimate shifts upward at the moment the model retracts and restarts a partial solution, a directional behavior no position-only predictor can reproduce (we note that this is qualitative, not aggregate). We frame this as approximate estimation of remaining generation length, distinct from exact-counting impossibility results for transformers, and interpret it as evidence that LLMs maintain a plan-like internal representation of output length (decodable, not necessarily used causally). Code: https://anonymous.4open.science/r/llm-output-length

Autonomous agents such as Claude Code and Codex now operate for hours or even days. Understanding their runtime behavior has become critical for downstream tasks such as diagnosing inefficiencies, fixing bugs, and ensuring better oversight. A primary way to gain this understanding is analyzing the reasoning trajectories and execution traces these agents generate. Yet such data remains in unstructured natural-language form, making it difficult for humans to interpret at scale. We introduce ACTONOMY (a combination of Action and Taxonomy, pronounced /ækˈtɑːnəmi/), a taxonomy for describing and analyzing agent behavior at runtime. has two components: (1) the taxonomy itself, developed through Grounded Theory and structured as a three-level hierarchy of 10 actions, 46 subactions, and 120 leaf categories; and (2) an open repository that hosts the living taxonomy, provides an automated analysis pipeline that applies it to agent trajectories analysis, and defines an extension protocol for community contributions. Our experiments show that ACTONOMY can compare behavioral profiles across agents and characterize a single agent's behavior across diverse trajectories, surfacing patterns indicative of failure modes. By providing a shared vocabulary, ACT*ONOMY helps researchers, agent designers, and end users interpret agent behavior more consistently, enabling better oversight and control.

Using prompted language models as classifiers enables classification in domains with limited training data, but misses some of the robustness and performance benefits that fine-tuning can bring. We study whether training on multiple classification tasks, each with its own prompt, improves performance on new domains with new classification prompts. We show that such training partially generalizes to adjacent domains, improving classification performance on tasks that are unseen during training. However, we identify specific edge cases where the finetuned models fail to follow prompts, such as when the classification prompt changes completely while the data domain remains the same as during training. We show that classification training can be mixed with general instruction following training, and that (when done well), such training keeps the benefits of classification training and mitigates its generalization failures. Surprisingly, we see that this no-thinking supervised classification training can generalize to with-thinking classification and summarization, suggesting that no-thinking classification training might be instrumentally useful in building other kinds of classifiers and monitoring systems.

Diffusion-based visuomotor policies perform well in robotic manipulation, yet current methods still inherit image-generation-style decoders and multi-step sampling. We revisit this design from a frequency-domain perspective. Robot action trajectories are highly smooth, with most energy concentrated in a few low-frequency discrete cosine transform modes. Under this structure, we show that the error of the optimal denoiser is bounded by the low-frequency subspace dimension and residual high-frequency energy, implying that denoising error saturates after very few reverse steps. This also suggests that action denoising requires a much simpler denoising model than image generation. Motivated by this insight, we propose Hydra-DP3 (${\bf HDP3}$), a pocket-scale 3D diffusion policy with a lightweight Diffusion Mixer decoder that supports two-step DDIM inference. Our synthetic experiments validate the theory and support the sufficiency of two-step denoising. Futhermore, across RoboTwin2.0, Adroit, MetaWorld, and real-world tasks, HDP3 achieves state-of-the-art performance with fewer than 1\% of the parameters of prior 3D diffusion-based policies and substantially lower inference latency.

Vector Quantization (VQ) based generative image compression has achieved remarkable perceptual quality. However, existing VQ codecs suffer from two fundamental limitations. First, they lack efficient content-adaptive entropy modeling and rely on static frequencies, leading to low coding efficiency. Second, the inherent conflict between discrete indices and continuous priors prevents true end-to-end joint Rate-Distortion (RD) optimization. To resolve these issues, we propose HyperVQ, a principled framework that establishes a high-performance hyperprior entropy foundation for VQ-based codecs. The core insight of HyperVQ is to shift probability modeling entirely into the continuous embedding space. Instead of directly predicting probabilities for discrete symbols, HyperVQ predicts a high-dimensional continuous multivariate Gaussian distribution for the continuous latents. By treating the discrete codebook entries as fixed "anchors" in this space, we convert the continuous Gaussian density into categorical index probabilities based on relative distances. This elegant formulation provides a powerful, spatially-adaptive entropy engine and renders the cross-entropy rate objective fully differentiable, empowering the network to actively and dynamically optimize the RD trade-off during training. To ensure practicality, we design the lightweight H Block and the Probability Estimation Engine (PEE) to facilitate highly parallel, millisecond-level inference. Experiments demonstrate that HyperVQ acts as a universal module across diverse VQ architectures (single-scale, large-codebook, RVQ), achieving an average bitrate saving of 18.5%, which is 7.28x the saving achieved by conventional Huffman coding. This establishes a robust, RD-controllable foundation for next-generation generative image compression.

The Allen--Cahn equation is one of the standard phase-field models for mean-curvature flow (MCF), parameterised by the diffuse-interface width $\varepsilon$. Practical simulations span a working range of $\varepsilon$, so a single model covering this range is needed. PDE-specific neural operators are accurate but must be retrained at every $\varepsilon$; general neural operators cover $\varepsilon$ in one model but lose the phase-field structure of the target. We introduce Interface-Augmented Diffusion-Reaction (IADR), a neural operator that addresses both shortcomings: a target network with a diffusion-and-reaction structure preserves the phase-field structure of the Allen--Cahn equation, while a hypernetwork that maps $\varepsilon$ to the weights of this target network amortises the dependence on $\varepsilon$ across the working range. A single IADR matches the per-$\varepsilon$ specialists on in-distribution data, retains their cross-shape generalisation under both $\varepsilon$- and geometry-out-of-distribution shifts, and runs at the per-$\varepsilon$ specialist's per-step inference cost. We further establish a $K$-step rollout error bound that aligns with the long-horizon behaviour observed in experiments.

Concept-based explanation methods have emerged as a prominent framework for interpreting deep neural networks using human-understandable concepts. However, existing approaches are limited to assigning a scalar importance score to a concept, failing to capture the interaction between the concept and its visual context. We argue that overlooking this interaction leaves a predictor's reasoning only partially understood. To address this gap, we introduce IDEA, a novel global post-hoc explanation framework that explicitly models the interaction between a concept and its visual context. Through this modeling, IDEA decomposes concept-level predictive information into three atomic components: uniqueness (independent concept contribution), redundancy (information shared with the context), and synergy (information arising strictly from concept-context interaction). Consequently, IDEA extends interpretability beyond merely identifying influential concepts to diagnosing how a predictor reasons -- for instance, flagging potential context dependencies that scalar scores cannot reveal. IDEA requires no task-specific concept annotations and no access to classifier internals. Evaluations on controlled and real-world settings demonstrate that IDEA can surface reasoning patterns invisible to traditional scalar concept scores.

Real-world spatial intelligence requires agents to understand scenes from continuous video streams, where objects move, persist, disappear, and reappear over time. While recent spatial foundation models have enabled generalizable feed-forward 3D reconstruction, most streaming methods remain geometry-centric and lack temporally consistent object-level understanding. Meanwhile, existing semantic reconstruction and 3D-aware vision-language methods largely rely on externally extracted 2D semantic cues or loosely coupled geometry inputs, limiting unified geometry-instance learning in long dynamic scenes. In this paper, we propose IGGT4D, a streaming instance-grounded geometry Transformer for online 4D scene understanding. IGGT4D processes video frames sequentially, reuses historical context through causal spatial-temporal modeling, and incrementally updates a unified representation of camera motion, geometry, and object identity. This enables long-sequence feed-forward reconstruction with geometry-instance consistency in dynamic environments. To address the lack of high-quality 4D supervision, we further construct InsScene4D-147K, a large-scale dataset spanning real/synthetic and static/dynamic scenes, with RGB images, depth, poses, and temporally consistent instance masks generated by an automated geometry-guided annotation pipeline. Experiments on 3D reconstruction, pose estimation, instance spatial tracking, and open-vocabulary segmentation demonstrate that IGGT4D outperforms existing streaming baselines while maintaining scalable online inference for long dynamic sequences.

We study the problem of learning in multi-player general-sum Markov games. In the simulator setting, where the learner can sample the next state conditioned on an arbitrary state action pair, the previous best-known upper bound of the sample complexity for learning an $\varepsilon$-approximate coarse correlated equilibrium (CCE) is $\widetilde{O}(H^4S\sum_{i=1}^m A_i/\varepsilon^2)$ (Li et al., 2022), where $H$ is the horizon, $S$ is the number of states, and $A_i$ denotes the number of actions for the $i$-th player. This work improves the upper bound to $\widetilde{O}(H^4S\max_{i\in[m]} A_i/\varepsilon^2)$, matching the lower bound of $\Omega(H^4S\max_{i\in[m]} A_i/\varepsilon^2)$ (Li et al., 2022). In the online setting, in which the learner can only sample a trajectory, the previous best-known sample complexity upper bound for learning CCE in multi-player general-sum Markov games is $\widetilde{O}(H^6S\max_{i\in[m]} A_i/\varepsilon^2)$ (Song et al., 2021; Jin et al., 2024; Mao et al., 2022). In this work, we improve this to $\widetilde{O}(H^5S\max_{i\in[m]} A_i/\varepsilon^2)$. To our knowledge, this is the tightest upper bound that breaks the curse of multi-agency. The core of our algorithmic design and analysis is the new gap decomposition and, in particular, a bandit algorithm with a high-probability empirical variance regret bound, which might be of independent interest.

Language brain-computer interfaces (BCIs), including speech and handwriting BCIs, hold significant promise for restoring communication in individuals with paralysis. However, the performance of current language BCIs remains limited due to the non-stationarity of neural activity. Existing neural decoding approaches typically treat neural non-stationarity as stochastic noise. For language BCIs, however, the noise is more complex due to the influence of semantic context in language sequences. For example, the articulation of the same phoneme can vary subtly depending on its surrounding content. Such context-induced variations introduce additional noise into neural representations and can confuse neural decoders, yet this issue has been largely overlooked in previous studies. Here, we propose explicitly modeling these context-induced noises to enhance neural decoding performance in language BCIs. Our key insight is that, unlike stochastic noise, context-induced noise carries useful information that can help the decoder disambiguate linguistically similar units. Therefore, by separating and modeling context-induced noise, we achieve improved decoding performance for language BCIs. Experiments on multiple datasets demonstrate that our approach achieves state-of-the-art performance.

Vision-language models (VLMs) learn strong statistical regularities during training, which can make them fail to perceive visual evidence when input images violate those regularities. Such failures are often treated as missing visual capability, but they may instead arise because the model has access to the relevant visual evidence and yet selects an answer dominated by learned training priors. In this work, we use controlled visual in-context learning to show that VLMs can overcome learned priors for visual tasks and use relevant visual evidence and capabilities. Specifically, we show that across five vision-centric tasks, demonstrations yield large gains on prior-conflicting counterfactual examples ($+27.2$% for Qwen3-VL 235B and $+18.1$% for Qwen3-VL 32B) while leaving real accuracy nearly unchanged. These findings also generalize across five VLM families with performance increases from $+9.7$% to $+30.3$%. To explain this effect, we analyze visual representations and find that demonstrations redirect attention toward grounding cues while reducing reliance on cues that support the learned prior. Finally, we find that in-context learning is limited when the underlying visual capability is weak or absent. Together, these results show that current VLMs possess visual grounding capabilities that standard evaluations may overlook, underscoring the need for controlled evaluation design in diagnosing and improving model capability.

Critical transitions – abrupt, often irreversible changes in system dynamics – arise across human and natural systems, often with catastrophic consequences. Real-world observations of such shifts remain scarce, preventing the development of reliable early warning systems. Conventional statistical and spectral indicators, such as increasing variance, tend to fail under realistic conditions of limited data and correlated noise, whereas existing deep learning classifiers do not extrapolate beyond their training data distribution. In this work, we introduce TipPFN, an in-context learning (ICL) framework that uses a prior-data fitted network to infer a system's proximity to a critical transition. Trained on our novel synthetic data generator, which is based on canonical bifurcation scenarios coupled to diverse, randomized stochastic dynamics, TipPFN flexibly capitalizes on contexts of various sizes, complexity and dimensionalities. We demonstrate robust, state-of-the-art early detection of critical transitions in previously unseen tipping regimes, sim-to-real examples, and real-world observations in both ICL and zero-shot settings.

Multimodal large language models must adapt to evolving tasks and domains, yet continual improvement under bounded deployment footprint remains difficult because repeated parameter updates or growing replay stores can accumulate adaptation state over time. We study fixed-footprint continual adaptation: the deployed adaptation state is kept under a fixed memory budget, while the backbone model is left unchanged and task-specific updates are externalized. We propose InduceKV, a retrieval-based method that stores each selected training prefix as an attention-ready memory entry, consisting of a frozen retrieval key and compact layerwise key--value (KV) payloads that can be appended to the model's self-attention cache. Under a strict memory budget, InduceKV constructs a compact inducing set through bilevel selection: a lightweight calibration is fit for retrieval, while the selected memory balances current-task likelihood, anchor-based retention, and coverage in the frozen retrieval space. Across task-incremental instruction tuning, continual VQA, domain-incremental adaptation, and lifelong multimodal instruction tuning, InduceKV consistently improves over PEFT, MoE, replay, and prompt-retrieval baselines under matched memory budgets. We further report backbone-matched, stage-1 CoIN, compute-matched, and scalability diagnostics, showing that the gains are not due to a stronger backbone, replay alone, or an unbounded candidate pool.

AI agents increasingly excel at generating, testing, and refining code. However, they fall short on tasks requiring formal guarantees of full coverage that testing alone cannot provide. Distributed systems are a prime example: properties such as consistency between reads and writes must hold under every possible interleaving of events. Mechanized formal verification can guarantee such correctness, but typically demands months to years of expert effort. As evidence, even SOTA coding agents (Claude Code with Opus 4.6, Codex with GPT-5.4) succeed on only 2/7 distributed key-value-store specifications. In this paper, we present the first effective approach to addressing this gap, Inductive Deductive Synthesis (IDS), which jointly and incrementally synthesizes implementation and proof, and learns from failed attempts to systematically try promising strategies. Built as an agentic LLM system, IDS achieves 7/7 in about 6.8 hours and $106 per spec on average, roughly 200× faster than expert effort and 17% cheaper than SOTA agents. IDS further incorporates performance feedback into the same loop, yielding implementations up to 3× faster than published verified systems.

Code generation and comprehension by Large Language Models (LLMs) have emerged as core drivers of industrial intelligence and decision optimization, finding widespread application in fields such as finance, automation, and aerospace. Although recent advancements have demonstrated the remarkable potential of LLMs in general code generation, existing benchmarks are mainly confined to single domains and languages. Consequently, they fail to effectively evaluate the generalization capabilities required for real-world industrial applications or to reflect the coding proficiency demanded by complex industrial scenarios. To bridge this gap, we introduce IndustryCode, the first comprehensive benchmark designed to span multiple industrial domains and programming languages. IndustryCode comprises 579 sub-problems derived from 125 primary industrial challenges, accompanied by rigorous problem descriptions and test cases. It covers a wide range of NAICS sectors, including Aerospace Product Mfg, Chemical Mfg, and Information, and incorporates diverse programming languages such as MATLAB, Python, C++, and Stata. In our evaluation, the top-performing model, Claude 4.7 Opus, achieved an overall accuracy of 71.9\% on sub-problems and 48.9\% on main problems. The benchmark dataset and automated evaluation code are now publicly available.

Many stochastic gradient methods are believed not to converge when the noise in stochastic gradients has only a finite $p$-th moment for $p\in\left(1,2\right)$, a setting known as the heavy-tailed noise assumption. However, some recent studies have found that Stochastic Gradient Descent ($\textsf{SGD}$), without any modifications to its update rule, can surprisingly converge in expectation for convex problems with bounded domains, highlighting the potential of classical stochastic gradient methods. Inspired by this recent progress, we provide a comprehensive study of stochastic optimization under heavy-tailed noise and establish new in-expectation convergence results for Stochastic Mirror Descent ($\textsf{SMD}$) in convex optimization and $\textsf{SGD}$ in nonconvex optimization. Notably, our results not only hold without algorithmic changes but also avoid restrictive assumptions such as bounded domains that were imposed in prior work. More importantly, our analysis provides a new, elegant, yet powerful framework for studying heavy-tailed stochastic optimization, opening a new route to understanding first-order stochastic gradient methods.

Neural responses to the same stimulus can vary dramatically with fluctuations in the brain's internal state and the organism's behavioral state. Yet studying how brain state shapes neural responses is difficult: limited experimental recordings sparsely sample this joint space, leaving many biologically valid state-stimulus pairings unobserved. Existing methods either capture state-dependent modulation with restricted model assumptions or synthesize neural activity without explicitly controlling internal brain state. We introduce BrainStateDiff, a conditional diffusion framework that models the spatiotemporal distribution of future neural activity and behavior given external stimuli and recent brain states. Motivated by two gaps in finite recordings, missing state-stimulus combinations and limited repeats of observed conditions, BrainStateDiff introduces two state-aware sampling strategies: cross-state sampling, which recombines observed states and stimuli to synthesize plausible neural responses from unobserved pairings, and matched-state sampling, which resamples from observed pairings to capture local trial-to-trial variability. We evaluate BrainStateDiff on our newly collected wide-field calcium imaging data in mouse V1, electrophysiology data in monkey V4, and two-photon calcium imaging data in mouse striatum. Our results show that BrainStateDiff generates biologically meaningful samples that preserve the state-dependent structure of real recordings and expand neural response space coverage beyond the original recordings. BrainStateDiff establishes state-conditioned generative modeling as a practical framework for studying how internal brain state shapes neural responses beyond the state-stimulus combinations directly observed in the experiments.

An extensive line of work studies fairness interventions for network embeddings, but less is known about their baseline behavior. In this work, we ask: how do baseline embeddings (without fairness interventions) produce disparate effects at the representation level? We analyze the asymptotic behavior of low-dimensional embeddings on stochastic block model (SBM) graphs, which encode both homophily and group structure. We characterize exact conditions under which embeddings cause information loss, showing that the amount of information loss depends directly on the graph’s density and assortativity. Notably, very different graphs can produce identical embeddings in the limit, and this non-invertibility disproportionately affects smaller and sparser communities. As a result, simple downstream tasks, such as link prediction, introduce higher error rates for these communities, helping explain disparities widely observed in practice.

Machine learning for powder X-ray diffraction (PXRD) symmetry classification is being trained at scale, yet the field lacks a common evaluation standard: papers report results on different subsets of the RRUFF mineral diffraction database, under different target taxonomies, with different preprocessing and different summary metrics, and published models are rarely released in a form that permits rescoring. We present InformedXRD, an evaluation framework with four parts: (i) a machine-readable mapping from 230 space groups to 99 extinction groups, the finest classes that powder diffraction can identify from systematic absences alone; (ii) two algorithmically curated real-data benchmarks, RRUFF-473 and RRUFF-325, whose inclusion rules are specified as code and reproduce from a frozen upstream snapshot; (iii) a prior-only frequency baseline and stratified reporting protocol that expose when apparent gains are driven by label imbalance or nuisance-fit severity; and (iv) a hierarchy-aware error metric on the condensed translationengleiche (same-lattice) subgroup graph that measures the crystallographic severity of misclassifications. Using four publicly released ViT checkpoints from Baggett et al. (2026) and an in-house 1D residual CNN as case studies, we show that evaluation design changes model ranking: standard paired tests point in different directions depending on whether one evaluates Top-1 or Top-5, and stratified analysis exposes a label-entropy confound that aggregate accuracy hides. All mapping tables, benchmark code, evaluation scripts, and case-study checkpoints are released as shared infrastructure for future PXRD-ML evaluation.

We study the instance-dependent sample complexity of one-dimensional stochastic bandit convex optimization in the fixed-confidence setting. First, we consider minimizer localization, where the learner seeks to identify a point within distance $\varepsilon$ of the unique minimizer. We introduce a new instance-dependent quantity $\Delta_\varepsilon(f)$, corresponding to the threshold for which the sublevel set of $f-\min f$ has length $2\varepsilon$. We prove a lower bound of order $\Delta_\varepsilon(f)^{-2}$ and give a bisection-style algorithm that matches this rate up to logarithmic factors. We then revisit the simple-regret setting, where the goal is to identify an $\varepsilon$-optimal point. Surprisingly, no comparable instance-dependent improvement is possible: every correct algorithm has sample complexity of order $\varepsilon^{-2}$ on every instance, matching the minimax rate up to constants. Together, these results separate the statistical complexity of minimizer localization from that of simple-regret optimization in one-dimensional convex bandits.

Continual learning seeks to acquire new knowledge while preserving previously learned representations. Despite the success of prompt-based methods, they remain fragile in domain-incremental learning scenarios, where shifts in the input distribution cause representation drift in shared layers and lead to forgetting. We introduce InTAct, a method that preserves functional behavior in shared representations without freezing parameters or retaining past data. InTAct constrains updates within activation regions associated with prior domains, while allowing flexible adaptation elsewhere, thereby stabilizing neuron functionality rather than directly restricting parameter values. The approach is architecture-agnostic, integrates seamlessly with existing prompt-based frameworks, and matches or outperforms state-of-the-art methods on domain-incremental benchmarks.

Language models are increasingly deployed as co-scientists in the real world, yet their ability to uphold research integrity, particularly under institutional pressures, remains unmeasured. We introduce IntegrityBench, a comprehensive benchmark evaluating three facets (misconduct classification, ethical action reasoning and artifact-grounded decision making), using 36 paired misconduct and ethical control tasks under a 5-level implicit-explicit pressure protocol across 3 domains and 4 research stages. Through a large-scale evaluation across 18 frontier model variants, we find that, under the strongest pressures, frontier models fail roughly one in three integrity-critical decisions. Surprisingly, neither scale nor reasoning can reliably enhance research integrity. Explicit pressures reliably induce compliance with misconduct while implicit contextual reframing more often causes over refusal of legitimate research tasks. Across tasks, models treat surface-level cues as evidence of misconduct without recognizing procedural justifications. Further, models failing to classify research requests accurately perform equally or better on artifact-grounded decision making (mean accuracy of 85.7 when classification fails versus 79.4 when it succeeds), demonstrating that the three facets might be structurally dissociated because accurate ethical actions don't require correct classification to precede them. Frontier models can appear helpful while displaying research integrity failures that create two distinct deployment risks: facilitating research misconduct and diminishing trust in AI-assisted research outputs.

Recent Vision-Language-Action (VLA) models have advanced end-to-end autonomous driving by incorporating reasoning for better interpretability and planning quality. However, most existing approaches directly generate the final trajectory without explicitly examining its future consequences, which limits their reliability in complex and dynamic environments. To address this limitation, we propose IRR-Drive (Intend, Reflect, Refine), an adaptive multimodal reflection framework for autonomous driving. Specifically, to tightly couple high-level reasoning with physical constraints, IRR-Drive first generates a preliminary textual intention and anticipates potential interactions by predicting future semantic bird's-eye view (BEV) representations. This dual-modality (Text + BEV) reflection space explicitly models anticipated scene evolution, enabling the model to rigorously self-correct and refine its initial intent before generating the final trajectory. Furthermore, to balance planning performance and computational efficiency, we construct reflection-oriented training data and design an adaptive reflection reward, enabling the model to adaptively select its reasoning mode according to scene complexity. Instead of using reasoning primarily as an auxiliary interpretation, IRR-Drive directly integrates an adaptive reflection mechanism into the planning framework, enabling grounded, decision-aware trajectory correction that is driven by scene complexity. Our method achieves state-of-the-art performance on the NAVSIM benchmark in both PDMS and EPDMS. Extensive experiments demonstrate the effectiveness of our multimodal reflection framework and validate the efficacy of the proposed adaptive reflection strategy.

Modern LLM serving systems use prefix caching to accelerate multi-turn conversational workloads. They cache the key-value states of previous conversation prefixes and reuse them when subsequent requests share the same prefix, to avoid redundant prefill computation and reduce serving latency. However, achieving a high cache hit rate remains challenging because cache reuse depends on whether a conversation continues and how soon the next turn arrives. Our analysis of real workloads shows that user intent is a strong signal for this behavior, as different intents exhibit different continuation probabilities and inter-turn delays. Based on this insight, we propose Intent-Aware Caching (IAC), an online eviction policy that classifies requests by intent, learns lightweight reuse statistics from recent serving logs, and uses them to guide an oracle-inspired eviction decision. Extensive evaluation on real workloads (e.g., ShareChat) shows that IAC reaches a 34.6\% cache hit rate, substantially outperforming the ML-based LPC (18.8\%) and LRU (12.2\%), while reducing average serving latency by up to 54.6\%.

Gaussian Process Variational Autoencoders (GPVAEs) effectively model sample dependencies via latent GP priors, but their inference remains computationally prohibitive at scale. Existing methods typically rely on inducing points, which are restricted to evaluations in the input domain and provide limited control over inductive bias. We introduce a scalable inference framework for GPVAEs that uses inducing variables defined as general linear functionals of the latent GPs, rather than point evaluations. Our formulation recovers the standard inducing-point GPVAE inference as a special case and yields a simplified reconstruction-minus-KL training objective that makes this connection explicit. More broadly, it provides a principled interface for incorporating operator-dependent representations and computational structures via the choice of inducing features. In particular, we instantiate this framework using Fourier features to capture global spectral correlations and B-spline features to induce efficient banded covariances. Across multiple tasks in representation learning, imputation, and conditional generation, our method offers a competitive accuracy-efficiency trade-off to existing approaches.

What if you could edit an image simply by drawing on it---circling an object, writing a short label, or sketching an arrow---with minimal or even no typed prompt? We first reveal a surprising capability: when a text-driven image editor is paired with a strong Vision-Language Model (VLM) at inference time, the combined system can directly interpret visual instructions embedded in the image itself, such as on-image text, bounding boxes, and directional arrows. However, this pipeline incurs additional VLM-planning latency, introduces brittle cross-model error propagation, and depends on external APIs or auxiliary model weights. To address these limitations, we introduce Siphon, a framework that internalizes this externally elicited visual-instruction-following competence into a single diffusion-based editor. Rather than relying on an external VLM at inference time, Siphon first uses a VLM planner to synthesize paired visual-instruction supervision, and then transfers this competence into the editor through lightweight LoRA fine-tuning. The resulting model reads, grounds, and executes on-image annotations directly from pixels, while keeping the diffusion sampling cost close to that of the base editor. Across multiple diffusion architectures, Siphon substantially improves spatial controllability, instruction adherence, and marker removal over text-driven baselines, while matching VLM-assisted pipelines without their additional planning latency, API dependence, and cross-model fragility. We do not position rendered long text as a replacement for conventional text prompts; rather, on-image text is mainly intended for short labels and spatially grounded edit intents, while long or complex instructions can still be provided through the standard text channel.

Mutual information (MI) is fundamental to representation-learning analysis in deep networks, but in high dimensions MI estimators depend on hyperparameters --- bin widths, kernel bandwidths, neighbourhood radii, auxiliary networks --- whose bias often dominates the signal. We show that continuous piecewise-linear (CPWL) networks, including ReLU networks, admit an intrinsic information-theoretic analysis that avoids this pitfall: their inherent partition of the input space into convex linear regions assigns to every input a discrete region label $\Pi$ alongside the continuous representation $T$. We decompose $I(Y; T)$ exactly into geometric terms, where the \emph{routing information} $I(Y; \Pi)$ admits a hyperparameter-free plug-in estimator that approximates a population lower bound on every other MI term in the decomposition. When the partition saturates, a functional-equivalence quotient at relative Frobenius tolerance $\varepsilon \in [0, 2]$ collapses regions implementing the same linear operator and restores informativeness. Empirically, the routing estimator correlates strongly with three established MI baselines {\it without any hyperparameter}, and the functional quotient at moderate $\varepsilon$ recovers informative estimates in the regime where the raw partition is bounded by $\log_2 N$ rather than $H(Y)$. We provide a parameter-free framework for analysing the internal information geometry of CPWL networks, grounded in their inherent piecewise-linear structure rather than in external discretisation.

Controllable 3D generation is a fundamental challenge in computer vision. Existing classifier-free guidance methods often yield suboptimal alignment with the given conditions, while Schr"{o}dinger Bridge approaches overlook intrinsic geometric discrepancies across multimodal manifolds, leading off-manifold trajectories and semantic misalignment. To address these issues, we propose Intrinsic-Preserving Schr\"{o}dinger Bridge (IPSB), a novel framework for direct part-aware 3D generation that leverages cross-modal intrinsic consistency. Specifically, we introduce Hierarchical Part-level Cross-modal Encoder that constructs hierarchical part-level representations through probability-based soft coverage, explicitly constraining graph topology to establish semantically consistent part graphs across modalities. Additionally, our IPSB incorporates asymmetric GW-inspired relational regularization, which strictly preserves the intrinsic semantic structure in the conditional manifold while allowing sufficient geometric flexibility in the 3D target manifold. Based on our IPSB framework, we devise Compositionally Controllable 3D Generation based on part-wise intervention inference, enabling independent replacement or resampling of target part nodes while fixing the trajectories of remaining components, thus achieving part-level controllable editing without full regeneration. Extensive experiments demonstrate that our superiority, efficiency and generality.

The achievement of practical fusion energy remains one of the most pressing unsolved scientific challenges, with immense implications for carbon-free power. A critical factor in the success of Inertial Confinement Fusion (ICF) is the design of a Laser Pulse Shape (LP) that can optimally drive implosions under stringent physical constraints. Traditional LP design relies on computationally expensive simulations and manual iterative refinement. We introduce the ICF Laser Pulse Shape Design System (LPDS), a generative inverse modeling framework that maps desired outcomes and target pellet configurations directly to optimized LPs. Crucially, we design a multi-objective loss function to ensure the generated LPs adhere to fundamental physical constraints and experimental feasibility. Furthermore, we present constraint-conditioning, inpainting, and gradient-based LP editing mechanisms for maximum fine-grained control over specific pulse characteristics during generation. Moreover, we validate our framework for LP design on a real-world experimental data. Our method establishes a data-driven inverse design framework for LP in ICF, contributing to the advancement of practical and sustainable fusion energy.

Recent multimodal large language models (MLLMs) achieve strong performance on reactive question answering, but real-world streaming assistants require proactive reasoning over continuous visual inputs. Existing benchmarks mainly study reactive or proactive interactions in isolated single-turn settings, overlooking dynamic multi-turn scenarios where users may add, modify, or cancel proactive requests alongside interleaved reactive queries. To address this gap, we introduce IPIBench, the first benchmark for evaluating Interactive Proactive Intelligence of MLLMs under streaming video settings. IPIBench covers proactive monitoring, proactive task management, and interleaved reactive–proactive requests. Evaluations on representative MLLMs reveal two major limitations: unstable proactive triggering and weak coordination between reactive and proactive behaviors. We further propose IPI-Agent, a training-free agentic framework with an interaction-control policy and a temporal-gating mechanism for stabilizing proactive triggering and coordinating multi-turn interactions. Experiments show that IPI-Agent consistently improves existing MLLMs across all benchmark settings.

Offline goal-conditioned reinforcement learning (GCRL) enables goal-reaching from static datasets, but the learned value, critic, or compatibility score must support both long-horizon reachability inference and offline policy extraction. We study iterative latent refinement as an architecture-level replacement for the feedforward value or critic backbone used by existing offline GCRL algorithms. Instead of emitting a score after a single feedforward computation, the network repeatedly updates a latent representation using shared recurrent update weights and lightweight step-specific parameters before exposing the final actor-facing signal. On the recent benchmark datasets reported in this study, recurrent variants improve performance across several algorithms without changing their losses, data, actors, or training protocol; across the 21 reported algorithm-dataset rows, the mean improvement is $10.7$ percentage points, with the largest gains on stitching and bottleneck maze tasks. CRL diagnostics are consistent with an improved policy-extraction interface: recurrent critics increase normalized separation between matched and mismatched goals and produce actors that remain closer to dataset actions under several behavior-support proxies. Depth and update-capacity studies further suggest that the effect depends on how computation is allocated between recurrent depth and per-step expressivity.

Universal Transformers (UTs) show strong performance on abstract reasoning tasks such as ARC-AGI, yet the specific sources of their performance gains remain underexplored. We conduct systematic ablations over UT variants and find that performance improvements on abstract reasoning are driven primarily by (i) the recurrent inductive bias induced by parameter sharing across depth and (ii) strong nonlinearity inside the Transformer block, rather than by elaborate architectural heuristics. Motivated by this finding, we propose the Universal Reasoning Model (URM). First, we propose ConvSwiGLU, which augments the feed-forward block with a channel-wise short convolution to effectively enhance nonlinearity of UT. Second, we adopt Truncated Backpropagation Through Loops (TBPTL) to eliminate noise from early loop when training with many loops, by freezing the gradients of early loops. Our approach substantially improves abstract reasoning performance, achieving 57.5% pass@1 on ARC-AGI 1 and 16.0% pass@1 on ARC-AGI 2. More importantly, we pretrain a 8-billion-parameter large language model based on the URM architecture from scratch, and the results show that URM still substantially outperforms the baseline across all reasoning tasks.

Contextual Integrity (CI) defines privacy not merely as keeping information hidden, but as governing information flows according to the norms of a given context. As large language models are increasingly deployed as personal agents handling sensitive workflows, adhering to CI becomes critical. However, even frontier models remain unreliable in making disclosure decisions, and existing mitigation strategies often degrade underlying task performance. To overcome this privacy-utility trade-off, we propose SelfCI, a complementary self-distillation framework that decouples information suppression from task resolution. SelfCI jointly optimizes two independent reverse KL divergences over distinct teacher distributions derived from feedback: one encourages preserving task-relevant information for utility, while the other enforces minimal and appropriate disclosure. This complementary formulation induces a Product-of-Experts (PoE) target, aligning the policy with the intersection of capability and privacy requirements. Without relying on costly external supervision, empirical evaluations demonstrate that SelfCI consistently outperforms competitive baselines such as online reinforcement learning algorithms (e.g., GRPO). These trends further extend to out-of-domain settings involving agentic workflows and accumulated private context, suggesting that SelfCI provides a practical path toward CI alignment.

We introduce JMed48k, a multi-profession Japanese healthcare licensing benchmark for evaluating vision-language models. Built from official PDF materials released by the Japanese Ministry of Health, Labour and Welfare, JMed48k contains 48,862 exam questions and 20,142 images from 11 national licensing examinations between 2005 and 2025, with visual content annotated under an 8-type taxonomy. From this corpus, we derive JMed48k-Eval, a recent five-year evaluation subset with 12,484 scored questions, including 9,905 text-only questions and 2,579 questions with images. We evaluate 21 proprietary, open-source, and medical-specific models, reporting text-only and with-image performance separately. Because these subsets contain different questions, we further introduce a paired image-removal audit that evaluates questions with images before and after removing visual content to explore four answer-transition states. The audit shows that proprietary and open-source models gain substantially from images, whereas medical-specific systems show limited observable use of visual evidence, with many correct answers persisting after image removal. Even among proprietary models, the net image-removal effect varies sevenfold across professions, from +5.7 points on Physician questions to +39.8 points on Public Health Nurse questions. We release JMed48k to support reproducible, profession-stratified evaluation of vision-language models in medical licensing settings.

Protein design generates amino acid sequences first and then translates them into DNA post-hoc. Control over the nucleotide sequence via codon choice at design time is thus limited. We introduce NuCaliby, a structure-conditioned model that jointly designs a protein's amino acid, mRNA, and DNA sequences at nucleotide resolution. It predicts a Potts model over a nucleotide graph, trained using amino acid supervision by marginalizing over synonymous codons. Its energy-based formulation supports composable inference-time guidance from both differentiable and non-differentiable objectives. NuCaliby preserves residue-level designability on par with Caliby while learning codon-aware nucleotide couplings directly from structure. Under organism-specific tRNA adaptation guidance, NuCaliby improves adaptation to host tRNA pools beyond synonymous sequence space while preserving protein designability. NuCaliby also embeds functional RNA motifs, directly into coding sequences, a problem that is combinatorially infeasible with extensive sampling alone. In wet lab experiments NuCaliby produces expressible and soluble proteins on par with state of the art methods, although it enables more complex nucleotide-level control and optimization. Together, these results extend protein design beyond the residue level, enabling joint optimization across the central dogma of molecular biology.

Training-free safeguards for text-to-image diffusion models often rely on a reusable safety signal, such as an unsafe direction or global toxic subspace, applied broadly across prompts. We provide a controlled geometric analysis of this global-unsafety assumption and reveal a consistent coverage-selectivity trade-off: compact unsafe subspaces fail to cover heterogeneous unsafe semantics, whereas broader aggregation increasingly distorts safety-adjacent benign prompts. Motivated by this finding, we propose CALM (Counterfactual Adaptive Local Modulation), a training-free safeguard that replaces uniform global removal with prompt-local counterfactual correction. Using matched unsafe-benign anchors, CALM routes each prompt to active unsafe categories, minimally edits only violating token representations toward the safe side, and suppresses positively aligned unsafe residual components. Across broad evaluation, CALM improves unsafe-content suppression while preserving benign utility, demonstrating that local counterfactual correction provides a more selective alternative to global unsafe-signal removal.

Post-LayerNorm Transformers have seen limited practical adoption due to enduring difficulties in scaling them to large depths. While prior research has focused on stabilising Post-LayerNorm by improving conditioning at initialisation, stability often deteriorates when models are trained with large learning rates, forcing additional compromises. In the decoder-only settings we study, existing Post-LayerNorm methods fail to outperform strong Pre-LayerNorm baselines. We propose KiteNorm, a novel normalisation method designed to break this trend. KiteNorm achieves stability through residual scaling and a regularisation technique based on hidden-state variance. Beyond stabilisation, KiteNorm learns separate scales for the skip and residual branches in each sublayer, improving performance. Across decoder-only Transformers up to 1B parameters, KiteNorm remains stable throughout training and consistently outperforms leading baselines, with scaling laws favouring KiteNorm across depth, width, batch size, and training budget. Ablations show that each component is necessary for the full stability and performance gains. We offer theoretical insights into KiteNorm through a new perspective on the gradient vanishing problem in Post-LayerNorm, linking the underlying hidden-state expansion to rank collapse.

Humans know when to reach for help e.g. $347 \times 28$ warrants a calculator while $2 + 2$ does not. Language models, by default, do not. Prompt-based approaches can instruct a model when to invoke tools, but this external scaffolding does not teach the model to recognize the boundary of its own knowledge. Reinforcement learning approaches that assign a single outcome reward to the whole trajectory fare no better: trajectory-level credit cannot isolate which tool call in a successful episode actually helped, nor penalize unnecessary calls. We propose \textbf{CARL} (\textbf{C}ompetence-\textbf{A}ware \textbf{R}einforcement \textbf{L}earning), which trains a critic on the model's own rollouts to learn where the model's parametric knowledge suffices and where it needs external help. By decomposing each rollout at natural tool-use boundaries (e.g., code fence delimiters and context block transitions), CARL assigns independent credit to each segment from a single binary outcome, without external judges or step-level annotations, addressing the credit-assignment limitations of trajectory-level methods. As a result, erroneous tool calls, incorrect extractions, and unnecessary calls each receive appropriately signed advantages under a single scheme. We show quantitatively and qualitatively that the trained critic captures the model's domain competence: it separates parametrically solvable from tool-dependent questions with AUC 0.93 at 7B. On five benchmarks spanning arithmetic, multi-hop factual QA, and numerical reasoning over financial tables, CARL improves exact-match accuracy by 6.7 points at 7B and 10.6 points at 3B over the strongest trajectory-level baseline (Search-R1 PPO), with the largest gain (+8.3 EM at 7B, +9.0 EM at 3B) on Musique, the most compositional multi-hop benchmark. Compared to a representative trajectory-level baseline (Search-R1 GRPO), the model issues 56\% fewer tool calls on questions answerable from parametric knowledge while remaining ${\sim}10$ EM points more accurate on those same questions, an emergent consequence of the critic learning where this model's competence ends. Gains are largest at small scale: the 3B model improvement is $1.6\times$ the 7B improvement, suggesting that knowing when to ask for help disproportionately benefits models with smaller parametric memory.

Memory enables long-context dialogue agents to maintain user state across many turns, but it also introduces a fundamental source-choice problem: grounding the response in the immediate context or retrieved prior memory? A wrong choice incurs two complementary costs: neglecting valid aged-out evidence, or overwriting sufficient context with stale memory. Existing systems and benchmarks largely conflate retrieval with source choice, relying on the model to resolve it implicitly and leaving this critical decision largely unexplored. In this paper, we introduce MSBench, a controlled benchmark that isolates source choice by pairing questions answerable exclusively from context with those requiring prior memory. Under this protocol, strategies that depend on the model's implicit judgment exhibit sharply degraded performance, exposing the limits of leaving source choice ungoverned. To this end, we propose the Memory-Source Reasoner (MSR), which frames source choice as an explicit metacognitive decision: it reasons about contextual sufficiency, answers directly when appropriate, and selectively retrieves missing evidence otherwise. Experimental results show that MSR achieves the best overall accuracy and selection accuracy across four answer backbones. On the primary GPT-5-mini run, it improves over the best non-MSR baselines by 12.6 points in overall accuracy and 6.3 points in selection accuracy. Our benchmark, code, prompts, and evaluation scripts are available at https://anonymous.4open.science/r/msbench-msr-0C72.

Continuous diffusion has served as a foundation for high-fidelity, controllable, and few-step generation across continuous data modalities such as images, videos, and molecular structures. However, in language modeling, prior continuous diffusion language models (DLMs) lag behind discrete counterparts. Existing categorical and simplex-based approaches operate over extremely large and sparse language spaces, while prior embedding-space approaches avoid this sparsity but lack a well-selected design space. In this work, we close this gap with LangFlow by connecting embedding-space DLMs to Flow Matching, alongside three key innovations: (1) we derive a novel ODE-based NLL bound for principled evaluation of continuous flow-based language models; (2) we propose an information-uniform principle for setting the noise schedule, which motivates a learnable noise scheduler based on a Gumbel distribution; and (3) we revise prior training protocols by incorporating self-conditioning, which improves both likelihood and sample quality for embedding-space DLMs and behaves differently from its use in discrete diffusion. Putting everything together, LangFlow is competitive with top discrete DLMs on both perplexity (PPL) and generative perplexity (Gen. PPL), reaching a PPL of 30.0 on LM1B and 24.6 on OpenWebText. It also exceeds autoregressive baselines in zero-shot transfer on 4 out of 7 benchmarks. As the first continuous DLM shown to rival discrete diffusion in both generative quality and perplexity, LangFlow provides clear evidence that continuous diffusion is a promising paradigm for language modeling. Code will be released upon acceptance.

Language-Assisted Image Clustering (LAIC) augments the input images with additional textual information with the help of vision-language models (VLMs) to improve clustering performance. Despite recent progress, existing LAIC methods mainly construct image-text pairs for text-modality utilization, which overlook two issues: (i) textual features constructed for images are highly similar, leading to weak inter-class discriminability; (ii) the training is restricted to pre-built image-text alignments, limiting the potential for better utilization of the text modality. To address these issues, we propose a new LAIC framework with two complementary components. First, we exploit cross-modal relations to generate more discriminative self-supervision signals for clustering, which is compatible with the pre-training mechanisms of VLMs such as CLIP. Second, we learn category-wise continuous semantic centers via prompt learning to produce the final clustering assignments. Extensive experiments on eight benchmark datasets demonstrate that our method achieves an average improvement of 2.7\% over state-of-the-art methods, and the learned semantic centers exhibit strong interpretability. Code is available in the supplementary material.

Fast linear algebra in deep learning usually comes with a choice: fixed geometry and exact computation, as in the Fourier transform, or adaptive geometry paid for by dense parameters, random features, or low-rank surrogates. To move beyond this trade-off, we introduce LAPLEX, a class of exact, trainable (phased) Laplace-kernel operators. A LAPLEX layer is a typically full-rank dense matrix, implicitly defined by learnable coordinate anchors, with FFT-like scaling. Consequently, it supports trainable matrix--vector operations at vector dimensions up to $10^9$ on modern GPUs. As a neural layer, it yields compact projections and classification heads interpretable as soft, trainable routing models. The same primitive also serves as an efficient Gram operator, enabling high-dimensional covariance models on flattened images of dimension $3 \cdot 10^6$ that preserve visible spatial structure without imposing convolutional bias. These applications reflect a single principle: dense geometry can be learned without storing a dense matrix, which enables data-adaptive global interactions in regimes where ordinary dense layers are out of reach. In this sense, LAPLEX separates expressivity from storage cost: it behaves like a dense trainable matrix, but is represented and applied through a small structured set of parameters.

Large language models increasingly serve as general-purpose reasoning engines, yet they remain unreliable on graph computation tasks whose inputs are discrete, permutation-equivalent, and algorithmic. We propose TRACER, a post-training framework that turns LLMs into native graph computational solvers. Our central thesis is that graph computation failures arise from mismatches within the Transformer computation itself: token order obscures graph symmetries, self-attention is biased toward textual proximity instead of graph topology, and autoregressive decoding weakly constrains algorithmic traces. TRACER addresses these challenges at three coupled levels. At the input level, we introduce permutation-invariant graph encoding, which encourages the LLM to map different serialization variants of a graph to a shared internal graph representation, thereby preserving graph symmetries. At the representation level, we enhance self-attention with the Topology-aware Residual Attention mechanism to inject graph topological signals into the attention kernel. At the reasoning level, we employ process-reward RL to encourage faithful execution of graph algorithmic traces. Across linear-time, polynomial-time, and NP-complete graph tasks, TRACER delivers substantial gains over baselines. These results suggest that equipping LLMs with graph-aware internal computation offers a practical path toward reliable neural solvers for structured algorithmic reasoning. The code is anonymously available here.

Large language model (LLM) agents often rely on long sequences of low-level textual actions, resulting in large effective decision horizons and high inference cost. While prior work has focused on improving inference efficiency through system-level optimizations or prompt engineering, we argue that a key bottleneck lies in the representation of the action space itself. We propose Latent Action Reparameterization (LAR), a framework that learns a compact latent action space in which each latent action corresponds to a multi-step semantic behavior. By reparameterizing agent actions into latent units, LAR enables decision making over a shorter effective horizon while preserving the expressiveness of the original action space. Unlike hand-crafted macros or hierarchical controllers, latent actions are learned from agent trajectories and integrated directly into the model, allowing both planning and execution to operate over abstract action representations. Across a range of LLM-based agent benchmarks, LAR significantly reduces the effective action horizon and improves inference efficiency under fixed compute budgets. As a consequence, our approach achieves substantial reductions in action tokens and corresponding wall-clock inference time, while maintaining or improving task success rates. These results suggest that action representation learning is a critical and underexplored factor in scaling efficient LLM agent inference, complementary to advances in model architecture and hardware.

While joint audio-visual understanding is fundamental to advancing machine cognition, current multimodal large language models (MLLMs) still struggle with complex cross-modal reasoning. Existing text-based chain-of-thought (CoT) compresses rich multi-modal features into discrete text, incurring information loss and inducing a language-bound phenomenon that diminishes attention to audio-visual signals. In contrast, a continuous latent space inherently preserves dense representations, serving as an ideal carrier for audio-visual information. Motivated by this, we propose $\textbf{LatentOmni}$, a novel cross-modal reasoning framework. By introducing a feature-level supervision mechanism to directly reconstruct raw sensory inputs within the latent space, LatentOmni leverages native latent features to bridge audio-visual modalities and text, ensuring sustained attention on original audio-visual inputs throughout reasoning. Furthermore, to maintain temporal consistency across modalities in latent space, we design Omni-Sync Position Embedding (OSPE), which generalizes multimodal rotary position encodings to drive audio-visual synchrony. To supervise this reasoning process, we construct LatentOmni-Instruct-35K, a dataset interleaving text with audio-visual segments that serve as dense evidence for latent reconstruction. Comprehensive evaluation across multiple audio-visual reasoning benchmarks demonstrates that LatentOmni substantially outperforms strong explicit-CoT baselines, validating latent space joint reasoning as a promising path toward genuine omnimodal understanding.

Geometric foundation models, such as the Visual Geometry Grounded Transformer (VGGT), provide strong 3D priors from unposed images. However, such models operate purely in a feed-forward, deterministic regime, \ie~they cannot generate plausible geometry beyond what the input views directly support. Generative models for 3D scenes, on the other hand, must rely on strong geometric priors to produce coherent outputs from sparse inputs. We bridge these two paradigms by performing flow matching directly in VGGT's latent space, leveraging its learned 3D priors without committing to any explicit downstream representation such as Gaussians, meshes, or video-VAE latents. This requires respecting the latent geometry: VGGT tokens occupy a product of high-dimensional hyperspheres on which standard Euclidean flow matching fails. We address this with a Riemannian Flow Matching framework defined on a product manifold of four hyperspheres, aligned with VGGT's multi-scale encoder, which keeps generated tokens on the valid data manifold required by the frozen decoding heads. On RealEstate10K and ScanNet++, our method achieves strong performance against recent scene generation baselines in both per-view 3D geometry and appearance, establishing latent-space flow matching on geometric foundation models as a viable paradigm for 3D generation.

Latent Diffusion Models (LDMs) typically adopt a two-stage pipeline: an auto-encoder (AE) is first pre-trained to define a latent space, then a diffusion model is trained to perform denoising within it. Such a two-stage design introduces a representation mismatch, as the latent space is optimized for reconstruction rather than adapting the denoising dynamics. We reveal that the LDM itself is intrinsically an AE, and consequently present LDM-is-AE, an end-to-end one-stage LDM training framework that eliminates the need for a separately trained tokenizer. Our key observation is that the LDM backbone actually performs a latent-to-feature-to-latent transformation at each denoising step, which can be interpreted as an internal decoding--encoding process. Leveraging this structure, we split the DiT backbone into two reciprocal components, DiT-E (i.e., DiT Encoding) and DiT-D (i.e., DiT Decoding), and impose image-space supervision on the intermediate features across all timesteps. Our model encourages the internal representation to align with the image domain throughout denoising, thereby establishing an explicit latent-to-image-to-latent path. At the zero-noise timestep, our model further performs an image-to-latent-to-image mapping, corresponding to an auto-encoding process. As a result, LDM-is-AE jointly learns latent representations and denoising dynamics in an end-to-end manner, yielding a diffusion-native latent space tailored to the generation process. Experiments demonstrate that LDM-is-AE exhibits highly competitive generation performance, achieving an FID of 1.84 on class-conditional image generation. Code and models will be released.

Arena-style rankings of language models are a widely used evaluation framework in leaderboards like LMArena, research papers, and model evaluation in production. These rankings aim to realistically assess model performance using pairwise comparisons of anonymous models on user prompts, aggregating votes via the Bradley-Terry model to produce the final ranking. However, we find that these leaderboards are highly vulnerable to several manipulation strategies that unscrupulous providers could use to boost a model's rank: strategic voting (individual votes that boost a model's rank, including votes on irrelevant models), strategic prompting (choosing prompts that favor a model), and strategic nomination (boosting the rank of a model by inserting irrelevant models). We demonstrate these effects on public leaderboard data, analyze the circumstances that cause these vulnerabilities, and describe approaches to safeguard against each attack. Notably, these issues hold for any pairwise preference-based evaluation that uses the Bradley-Terry model. These manipulations boost the rank of nearly every tested model; each attack can boost a model by up to 15-21 places in a ranking and, combined, they can boost a model by up to 45 places.

Recently, Antoniadis et al. (ICLR 2025) proposed a framework for incorporating predictions to approximate NP-hard selection problems. Despite its simplicity, this approach tightly matches theoretical lower bounds, making its generalization highly compelling. We address an open question raised in the work of Antoniadis et al., concerning the extension of this approach to other important problems outside the class of selection problems, such as scheduling. We develop a learning-augmented algorithm for the makespan minimization problem on unrelated machines, denoted by $R||C_{\max}$. By using predictions of heavy job assignments, we achieve a polynomial-time $(1+\varepsilon)$-approximation for accurate predictions that smoothly degrades to a worst-case 2-approximation as the error increases. We conclude our work with an empirical analysis of our method.

Command line interface (CLI) agents are emerging as a practical paradigm for agent-computer interaction over evolving filesystems, executable command line programs, and online execution feedback. Recent work has used reinforcement learning (RL) to learn these interaction abilities from verifiable task feedback, yet few methods exploit the native structured attributes of CLI actions as learning signals. Beyond this underused action structure, CLI learning also couples two bottlenecks for coding agents. First, the agent must identify task-relevant evidence in a large codebase from partial observations. Second, sparse terminal rewards must be assigned to the actions that shape a long multi-turn trajectory. We study these bottlenecks through shell-driven information extraction and file editing tasks. For selective observation, we introduce $\sigma$-Reveal, an inference-time mechanism that selects token-budgeted context for the same CLI. For credit assignment, we propose Action Advantage Assignment ($\mathrm{A}^3$), a native agentic RL method that preserves the algorithmic complexity of standard agentic RL. $\mathrm{A}^3$ constructs turn-level advantages from episode-level relative feedback, abstract syntax tree (AST) based action sub-chain residuals, and tree-level trajectory margins. To further evaluate this problem setting, we construct ShellOps, a verifiable dataset suite covering CLI tasks in repository environments.

Learning a response law for continuous decisions from observational time series is central to many real-world decision systems, yet temporal non-stationarity poses a fundamental challenge: outcome levels drift over time, causing naive models that fit raw outcomes to conflate treatment effects with shifting baseline conditions. Even when the one-step response is identifiable, unrestricted cross-time generalization remains fundamentally ill-posed. Thus, we introduce a constrained decision target based on a causal action geometry with anchored link-scale morphologies. Our approach constructs anchored contrasts over a family of link functions, isolating the action-dependent component of the response that can be shared across periods while allowing flexible, period-specific calibration. This yields a morphology-restricted oracle target for continuous decision making under temporal drift. Building on this formulation, we develop a two-stage orthogonal estimator that first obtains cross-fitted pilot estimates of anchored responses and then performs honest profiled selection over morphology and representation classes. The resulting estimator admits oracle guarantees for the restricted target and, under approximate persistence, produces decision rules that remain near optimal in future periods. Experiments on synthetic data and a large-scale e-commerce pricing application support the proposed target and its decision benefits under temporal shift.

Solving inverse problems on video requires priors over both frame appearance and temporal dynamics. Many recent approaches combine pretrained image diffusion priors with motion guidance, sidestepping the costs of training large video models. Temporal consistency is often enforced via photometric losses or warping-based constraints, which implicitly assume that motion explains frame-to-frame or latent-to-latent changes up to small deviations, an assumption that fails under occlusions, lighting changes, and non-rigid deformation. In this paper, we propose to learn the coupling between motion and frame appearance as a generative prior. We learn the distribution of motion-induced appearance residuals with a conditional diffusion model and use it to regularize reconstruction in video inverse problems. Our motion-appearance prior acts as a plug-in regularizer for diffusion-based video reconstruction, preserving the benefits of image diffusion models while remaining compatible with existing appearance regularization techniques. Experiments on dynamic video inverse problems show that, depending on the task, reconstruction with a learned motion-appearance coupling prior matches or outperforms feature-consistency and noise-warping baselines.

Reinforcement learning from verifiable rewards (RLVR) produces strong reasoning models, yet these models can fail catastrophically when the conditioning context is fallible (e.g., corrupted chain-of-thought, misleading partial solutions, or mild input perturbations), because standard RLVR optimizes final-answer correctness only under clean conditioning. We introduce GASP (Guided Adversarial Self-Play), a robustification method that explicitly trains error detection and repair capabilities using only outcome verification. Without human labels or external teachers, GASP instantiates an adversarial self-play game within a single model: a polluter learns to induce failure through locally coherent corruptions, while an agent learns to diagnose and recover under the same corrupted conditioning. To address the scarcity of successful recoveries early in training, we propose in-distribution repair guidance, an auxiliary imitation objective on self-generated repairs that increases recovery probability while preserving previously acquired capabilities. Across four open-weight models (1.5B–8B), GASP converts strong-but-brittle reasoners into substantially more robust ones that better withstand misleading and perturbed context, while often improving clean accuracy. Further analysis shows that adversarial corruptions create an effective curriculum, and that in-distribution guidance enables rapid recovery learning with minimal representational drift.

Fine-grained image retrieval (FGIR) typically relies on supervision from seen categories to learn discriminative embeddings for retrieving unseen categories. However, such supervision often biases retrieval models toward the semantics of seen categories rather than the underlying appearance characteristics that generalize across categories, thereby limiting retrieval performance on unseen categories. To tackle this, we propose GAPan, a Generative Appearance Prior alignment network that reformulates the learning objective from category prediction toward appearance modeling. Technically, GAPan treats retrieval features with an invertible density model based on normalizing flows. In the forward direction, the flow maps all instance features into a latent density space, where each seen category is modeled by a class-conditional Gaussian prior and optimized via exact likelihood estimation. This formulation preserves richer appearance details by leveraging the invertible property of the flows. In the reverse direction, samples from the high-density regions of these learned priors are mapped back to the feature space to produce appearance-aware anchors that reflect intra-category variation. These anchors supervise a prior-driven alignment objective that aligns retrieval embeddings with category-specific appearance distributions, thereby improving generalization to unseen categories. Evaluations demonstrate that our GAPan achieves state-of-the-art performance on both widely-used fine- and coarse-grained benchmarks.

The statelessness of foundation models bottlenecks agentic systems’ ability to continually learn, a core capability for long-horizon reasoning and adaptation. To address this limitation, agentic systems commonly incorporate memory modules to retain and reuse past experience, aiming for continual learning during test time. However, most existing memory designs are human-crafted and fixed, which limits their ability to adapt to the diversity and non-stationarity of real-world tasks. In this paper, we introduce ALMA (Automated meta-Learning of Memory designs for Agentic systems), a framework that meta-learns memory designs to replace hand-engineered memory designs, therefore minimizing human effort and enabling agentic systems to be continual learners across diverse domains. Our approach employs a Meta Agent that searches over memory designs expressed as executable code in an open-ended manner, theoretically allowing the discovery of arbitrary memory designs, including database schemas as well as their retrieval and update mechanisms. Extensive experiments across four sequential decision-making domains demonstrate that the learned memory designs enable more effective and efficient learning from experience than state-of-the-art human-crafted memory designs on all benchmarks. When developed and deployed safely, ALMA represents a step toward self-improving AI systems that learn to be adaptive, continual learners.

Reinforcement Learning with Verifiable Rewards (RLVR) enhances large language model (LLM) reasoning, yet growing evidence indicates an \emph{exploration ceiling}: they tend to reweight existing solution traces and produce low-diversity rollouts. This restricts the discovery of novel solution perspective and weakens out-of-distribution generalization. We address this limitation by directly perturbing the rollout policy's parameters while updating the clean policy model as usual, an approach that achieves better temporal consistency and yields more complex behavior patterns. To stabilize learning from these perturbed rollouts, which are inherently off-policy, we incorporate truncated importance sampling. Furthermore, we introduce a lightweight, dynamic noise scheduler based on semantic diversity and model self-certainty to optimally adjust the noise scale. Instantiated on GRPO, our method (PSN-GRPO) demonstrates its efficacy through four key contributions: (1) it improves high-budget pass@$k$, increases semantic and operational diversity, increases CoT consistency, and remains complementary to exploration-oriented RLVR techniques; (2) it validates efficacy across multiple models, including Qwen2.5, Qwen3, and Llama-3.1; (3) it shows robust performance on out-of-distribution science and coding benchmarks; and (4) it uncovers novel solution perspectives for problems the original model could not solve, as shown by qualitative analysis.

On-policy distillation (OPD) has emerged as an efficient post-training paradigm for large language models. However, existing studies largely attribute this advantage to denser and more stable supervision, while the parameter-level mechanisms underlying OPD's efficiency remain insufficiently understood. In this work, we argue that OPD's efficiency stems from a form of ``foresight'': it establishes a direct and stable path from the initial model to the final model early in training. This foresight manifests in two aspects. First, at the \textbf{Module-Allocation Level}, OPD identifies regions with low marginal utility and concentrates updates on modules that are more critical to reasoning. Second, at the \textbf{Update-Direction Level}, OPD exhibits stronger low-rank concentration, with its dominant subspace aligning closely with the final update subspace early in training. Motivated by this theory, we propose \textbf{EffOPD}, a plug-and-play acceleration method that speeds up OPD by searching for an effective step size and extrapolating along the current update direction. EffOPD requires no additional trainable modules or complex hyperparameter tuning, and achieves up to $3.5\times$ training acceleration while maintaining comparable final performance. Overall, our findings provide a parameter-dynamics perspective for understanding the efficiency of OPD and offer practical insights for designing more efficient post-training methods for large language models. Our code is available at: https://anonymous.4open.science/r/EffOPD-7C58.

Text-to-image diffusion models fail to generate correct object counts in dense scenes, where overlapping instances collapse into indistinguishable structures despite appearing visually plausible. We identify this as instance ownership collapse: tokens from overlapping objects interact freely through attention, while heavily occluded instances receive weak supervision due to their small visible areas. We address this through layout-aware attention biases that softly bias token interactions toward region-consistent grouping and suppress cross-instance leakage, paired with an amodal-balanced loss that amplifies gradients for occluded objects based on their occlusion level. To enable systematic evaluation, we introduce OverlapDepth-45K, a benchmark of densely overlapping scenes with amodal supervision. Our approach substantially improves count accuracy and prevents instance merging while preserving image quality.

We study infinite-horizon average-reward constrained Markov decision processes (CMDPs) under the weakly communicating assumption. Our contributions are twofold. First, we establish strong duality for weakly communicating average-reward CMDPs over stationary policies with finite state and action spaces. Despite the absence of a linear programming formulation and the resulting nonconvexity under the weakly communicating setting, we show that strong duality still holds by carefully exploiting the geometric structure of the occupation measure set. Second, building on this result, we propose a primal--dual clipped value iteration algorithm for learning weakly communicating average-reward linear CMDPs. Our algorithm achieves regret and constraint violation bounds of $\widetilde{\mathcal{O}}(T^{2/3})$, improving upon the best known bounds, where $T$ denotes the number of interactions. Our approach extends clipped value iteration to the constrained setting and adapts it to a finite-horizon approximation, which stabilizes the dual variable and is crucial for achieving improved regret bounds. To analyze this, we develop a novel approach based on strong duality that enables the decomposition of the composite Lagrangian regret into separate bounds on regret and constraint violation.

Multi-agent systems have emerged as a promising paradigm for multimodal medical reasoning, but existing methods rely on two assumptions that hinder practical deployment: (i) always-active dense collaboration among multiple agents, and (ii) dependence on large cloud-based models that raise privacy and efficiency concerns. In this paper, we challenge both assumptions and propose SMART-Med, a fully local multi-agent method that treats collaboration as a learnable decision rather than a default behavior. Our key insight is that not every medical query requires costly multi-agent deliberation: many can be reliably solved by a single well-trained agent, while only uncertain cases benefit from expert collaboration. Based on this insight, SMART-Med introduces an uncertainty-driven routing mechanism that adaptively decides when to answer directly and when to recruit a small subset of expert agents, effectively converting dense multi-agent reasoning into selective, query-adaptive coordination. To make this routing reliable, we curate a new multimodal medical dataset MedVR-3K with high-quality reasoning traces and a difficulty hierarchy, and then fine-tune local agents using GRPO with two complementary rewards: a certainty-aware accuracy reward and a vision-language reward model for correctness and reasoning quality respectively. Experiments on four medical visual question answering benchmarks show that SMART-Med achieves competitive performance compared to existing MAS approaches while substantially reducing inference cost (reducing inference time by more than 3.8$\times$ and token usage by 72\%). Moreover, after fine-tuning on MedVR-3K, SMART-Med yields an average improvement of 6.7\% across Slake, PathVQA, VQA-RAD, and PMC-VQA. Our results suggest that selective collaboration can be an efficient yet effective alternative to dense collaboration and enable practical, privacy-preserving deployment of multi-agent medical AI.

Long-range geophysical forecasts are fundamentally limited by chaotic dynamics and numerical errors. While data assimilation can mitigate these issues, classical variational smoothers require computationally expensive tangent-linear and adjoint models. Conversely, recent efficient latent filtering methods often enforce weak trajectory-level constraints and assume fixed observation grids. To bridge this gap, we propose Latent Ensemble Variational Data Assimilation (LEVDA), an ensemble-space variational smoother that operates in the low-dimensional latent space of a pretrained differentiable neural dynamics surrogate. By performing four-dimensional ensemble-variational (4DEnVar) optimization within an ensemble subspace, LEVDA jointly assimilates states and unknown parameters without the need for adjoint code or auxiliary observation-to-latent encoders. Leveraging the fully differentiable, continuous-in-time-and-space nature of the surrogate, LEVDA naturally accommodates highly irregular sampling at arbitrary spatiotemporal locations. Across three challenging geophysical benchmarks, LEVDA matches or outperforms state-of-the-art latent filtering baselines under severe observational sparsity while providing more reliable uncertainty quantification. Simultaneously, it achieves substantially improved assimilation accuracy and computational efficiency compared to full-state 4DEnVar.

We introduce LibriBrain100, a large-scale MEG dataset for speech decoding designed from the ground up for reproducible, standardised evaluation. LibriBrain100 more than doubles the size of the original LibriBrain release, resulting in over 100 hours of high-quality MEG acquired while subjects listened to naturalistic continuous speech. With ~80 hours from a single subject, LibriBrain100 sets a new record for ‘deep’, within-subject neural data (8x more than the next comparable dataset and roughly 80x more than other datasets). To demonstrate the payoff of this depth-first design, we evaluate on a word-classification benchmark—an increasingly well-established stepping stone towards the open challenge of non-invasive brain-to-text decoding. Using an existing decoding model, we achieve state-of-the-art performance—validating both the quality of the recordings and the value of within-subject data at scale. Because collecting 80 hours of data per user is impractical for real-world applications, we also collected ~40 minutes of additional data from each of 32 subjects. Using the same word-classification benchmark, we demonstrate the value of ’broad’ multi-subject data: supervised fine-tuning of a pre-trained model can substantially compensate for limited per-subject data. We provide standard train, validation, and test splits, all reproducible through an open-sourced Python library that supports easy downloading, optional preprocessing, and data loading for common deep learning frameworks. In addition, the dataset and evaluation infrastructure are being released alongside an open machine-learning competition with a public leaderboard for standardised benchmarking. Ultimately, our hope is that LibriBrain100 will accelerate progress towards practical non-invasive brain-computer interfaces, capable of restoring communication to people living with severe paralysis.

Scaling large recommendation systems requires advancing three major frontiers: processing longer user histories, expanding candidate sets, and increasing model capacity. While promising, transformers' computational cost scales quadratically with the user sequence length and linearly with the number of candidates. This trade-off makes it prohibitively expensive to expand candidate sets or increase sequence length at inference, despite the significant performance improvements. We introduce **LIME**, a novel architecture that resolves this trade-off. Through two key innovations, LIME fundamentally reduces computational complexity. First, low-rank ``link embeddings" enable pre-computation of attention weights by decoupling user and candidate interactions, making the inference cost nearly independent of candidate set size. Second, a linear attention mechanism, **LIME-XOR**, reduces the complexity with respect to user sequence length from quadratic ($O(N^2)$) to linear ($O(N)$). Experiments on public and industrial datasets show LIME achieves near-parity with state-of-the-art transformers but with a 10$\times$ inference speedup on large candidate sets or long sequence lengths. When tested on a major recommendation platform, LIME improved user engagement while maintaining minimal inference costs with respect to candidate set size and user history length, establishing a new paradigm for efficient and expressive recommendation systems.

LiDAR relocalization based on scene coordinate regression (SCR) degrades sharply when test trajectories traverse route segments absent from training, a practical challenge in long-range autonomous-driving deployment, where complete route coverage is rarely achievable. We present LINK, a confidence-guided framework for relocalization under partial scene coverage. LINK predicts an absolute pose hypothesis and scene features with an SCR backbone, estimates pose reliability from temporal pose-feature sequences, and uses a decision policy to arbitrate between direct absolute hypotheses and relative motion propagated from reliable historical estimates, thereby preserving globally referenced localization across uncovered segments. We also introduce UrbanUnseen, a city-scale benchmark with native known-to-unknown scene transitions, together with controlled partial-coverage protocols on Oxford and NCLT. Experiments show that LINK substantially improves robustness when training coverage is incomplete, consistently outperforms representative map-free baselines under partial coverage, and remains effective in fully known scenes.

Importance sampling provides an elegant variance-minimization principle for stochastic optimization: sampling examples proportional to their gradient norms yields a minimum-variance gradient estimator. However, in modern deep learning, importance sampling often fails to outperform uniform sampling once proxy-computation cost, stale scores, and fair update budgets are accounted for. We propose LITE, a lightweight lazy sampler for practical importance sampling. LITE maintains inexpensive proxy scores, updates them only periodically, smooths them across refreshes, and converts them into sampling probabilities through a temperature rule that prevents over-concentration. When proxy scores are informative, LITE prioritizes high-impact examples; when they are noisy or stale, the sampling distribution moves back toward uniform. We analyze the resulting variance reduction in terms of proxy quality, temporal drift, and temperature, yielding an explicit condition for when LITE improves over uniform sampling. Experiments under matched update budgets and end-to-end wall-clock accounting show that LITE is competitive with or better than uniform sampling across vision, text, and 3D benchmarks, with the clearest gains in scarce-budget and imbalanced regimes.

Diffusion personalization deployments serve growing catalogues of LoRA adapters, but resident-LoRA serving is capped near the GPU memory budget (about 150 rank-16 adapters on an A100 80 GB GPU running FLUX.1-dev). We introduce LITHE, a LoRA-compatible deployment representation that stores each adapter as an integer-index stream over a deterministic Hadamard codebook, and that pushes this resident-adapter ceiling to a verified 10,000 adapters on the same single GPU at sub-80 GB peak. The same on-disk indices drive two serving modes: an index-resident mode that routes a heterogeneous batch of adapters in one forward, and a decoded compiled mode within 1% of an optimized LoRA+compile baseline at 28-step inference at 1024² resolution on FLUX.1-dev (and 1.07–1.43× faster at lower resolutions). On disk, each rank-16 FLUX.1-dev adapter occupies 1.81 MB on average across the benchmark pool, a 50×/200× reduction over the matched r=16 / r=64 LoRA disk payload after the same coder.

Interactive streaming music generation promises the use of generative models for live performance and co-creation that is impossible with offline models. However, SOTA models exist in the discrete-AR regime, requiring industrial levels of compute for both training and inference. In this work, we investigate whether audio diffusion models, with their wide support in the open-source community but non-streaming bidirectional nature, can be repurposed efficiently into interactive models accessible on consumer hardware. By taking a critical look at the modern pipeline for block-wise outpainting diffusion, we identify critical inefficiencies during inference that result in strictly worse computational efficiency than their discrete-AR counterparts. We propose Live Music Diffusion Models (LMDMs), a simple modification of the generative diffusion process that recovers, and then outperforms, the inference complexity of the discrete Live Music Models (LMMs) through block-wise KV Caching. Unlike LMMs, LMDMs further enable stable post-training alignment through our novel ARC-Forcing paradigm, reducing error accumulation without any explicit RL or reward models. We demonstrate the application of LMDMs in a number of creative domains, including text-conditioned generation, sketch-based music synthesis, and jamming. We finally show how LMDMs can be used as a generative instrument in a real artist-AI collaboration, utilizing LMDMs as a “generative delay” to transform musicians’ improvisation live for variable timbral effects while running locally on a consumer gaming laptop.

Confidence-weighted routing, selective abstention, and ensemble weighting all assume that a model's stated confidence is informative about its capability on the question being asked. They presume functional metacognition, the capacity to assess one's own capabilities, without exercising them, relative to other frontier language models. Aggregate calibration is well studied, with mixed results, but the underlying structure of elicited confidence is less well understood. We decompose binary confidence judgements from 20 frontier Large Language Models (LLMs) across six benchmarks using tetrachoric factor analysis paired with pairwise calibration, asking whether two models that differ in confidence also differ in performance. On factual recall and information retrieval benchmarks the cross-model confidence matrix is approximately rank-one and a single dominant factor captures most of the latent variance. Models retrieving facts share an item-level difficulty axis and differ mainly in their decision thresholds along it. Across all benchmarks the relationship between confidence and performance collapses once items that all models agree on are removed. Inter-model pairwise calibration is small even where statistically significant, and what remains shrinks to nothing once base-rate differences along the shared factor are controlled for. Mathematical reasoning is the apparent exception, but this turns out to be a confound where reasoning models answer questions about their confidence by trying to solve them in their chain of thought, bypassing the sub-symbolic self-knowledge we seek to measure. We find no evidence for significant verbalised individuated metacognition in any tested domain. A trivial logistic regression on surface features of the confidence-judgement reasoning text beats the median confidence model's repeatedly sampled judgements on half the benchmarks, evidence that extracting uncertainty from the reasoning trace recovers signal the binary judgement does not.

Membership inference attacks (MIAs) are commonly formulated as single-sample hypothesis tests with false positive rate control. However, realistic adversaries often test many candidate samples and aggregate the declared members, making membership inference a multiple testing problem. In this regime, per-sample calibration can produce unreliable sets of inferred members. We develop a false discovery rate (FDR)-controlled framework for membership inference based on the classical local FDR formulation. We instantiate the framework for high dimensional ridge regression. Under a conditional sufficiency condition, we establish FDR guarantee of the proposed method, and derive an asymptotic characterization of its detection power in high-dimensional regimes under an isotropic Gaussian design. Our analysis shows that, under FDR control, stronger ridge regularization reduces membership inference risk. Experiments on synthetic and a real-world dataset validate the theoretical findings and demonstrate that the proposed methods provide more reliable membership discoveries than existing attacks.

While Proximal Policy Optimization (PPO) demonstrates strong performance in stationary settings, we show that its standard optimization paradigm struggles in continual and non-stationary environments. The failure does not stem from insufficient model capacity or overly restrictive clipping. Instead, PPO performs persistent, directionally inefficient local updates, which indicates a lack of geometry-aware guidance for accumulating meaningful behavioral change and ultimately hindering transitions toward new behavior patterns. Although divergence-based regularization introduces partial geometric awareness, its monotonically increasing penalties implicitly discourage large policy deviations, even when such shifts are necessary for effective adaptation. To address this limitation, we propose Gaussian Trust Region Policy Optimization (GTR), which reshapes the trust region using a Gaussian kernel. The resulting constraint is bounded and non-monotonic, providing strong local stability while progressively relaxing under sustained high-advantage updates. To further improve robustness, we introduce a Mixture Gaussian Anchor that adapts to recent policy trajectories, reducing variance induced by stale references. GTR is architecture-agnostic and achieves strong performance across games, simulated robotic control, open-world exploration, and language model post-training. These results demonstrate that geometry-aware trust-region design can be a promising direction for robust reinforcement learning in complex non-stationary environments.

Multi-objective reinforcement learning (MORL) aims at optimising several, often conflicting goals to improve the flexibility and reliability of RL in practical tasks. This is typically achieved by finding a set of diverse, non-dominated policies that form a Pareto front in the performance space. However, constructing such a policy set remains computationally demanding, as it requires finding not a single optimal policy, but a set of policies to cover a wide range of trade-offs among the objectives. We introduce LLE-MORL, an approach that traces policy manifolds by utilising the local relationship between the high-dimensional policy parameter space and the performance space. This structured representation enables an efficient search within contiguous solution domains via locally linear extrapolation, allowing for the rapid generation of high-quality solutions without extensive retraining. We also provide a theoretical analysis of the policy manifold structure in order to specify the applicability of the method. Experiments across diverse continuous control domains demonstrate that LLE-MORL consistently achieves higher Pareto front quality and efficiency than state-of-the-art approaches.

Deployment-time safety methods for large language models (LLMs) are predominantly supervised and assume access to unsafe training data. Nevertheless, new attacks and harm categories regularly arise, not captured by models trained in such a supervised fashion. An alternative approach is to view this problem through the lens of anomaly detection, namely, to rely solely on modeling safe data and flagging out-of-distribution inputs. However, LLM activations lie in a high-dimensional space, raising concerns about whether anomaly detection is statistically feasible. We show that, under the linear representation hypothesis (LRH), there may indeed be hope. In the LRH concept space, which is typically recovered via a sparse autoencoder (SAE), nearby points share a small common active support. Based on this local sparsity insight, we propose a framework for locally masked SAE-based anomaly detection, and establish a sample complexity bound that scales polynomially in the local sparsity and only \emph{logarithmically} in the SAE's ambient dimension. Finally, we empirically validate our framework on various architectures and datasets, including both capability-testing datasets and safety-specific datasets.

Open-vocabulary multi-label recognition (OV-MLR) aims to identify all queried semantic concepts present in an image. Existing methods mainly improve global image-text matching or text-side representations, leaving local prediction largely under-explored. In this paper, we revisit local prediction and reveal that it can serve as a strong discriminative branch once two properties are properly handled: spatial consistency of local visual features and compatibility-aware local competition. This perspective is motivated by local semantic sparsity: each image region is related to only a small subset of queried concepts, while compatible concepts of different granularities or types may still coexist locally. We instantiate this perspective with LoCo, a selective Local Competition framework for OV-MLR. LoCo adopts a spatially consistent visual encoder to reduce local feature entanglement, and introduces Selective Softmax to impose competition only among locally incompatible categories while preserving compatible responses. It further combines multi-scale local aggregation with global prediction to handle objects at different scales and scene-level semantics. Experiments on four multi-label benchmarks show that LoCo achieves state-of-the-art performance, with average gains of +11.3\% F1-score and +5.2\% mAP. Our results suggest that local semantic sparsity offers a useful basis for developing more discriminative open-vocabulary multi-label recognition methods. Code is available in the supplementary material.

Aligned language models refuse unsafe requests because RLHF widens the logit margin between refusal and affirmative tokens at the first decoded position. We call this scalar the refusal-affirmation logit gap and use it as a per-prompt diagnostic for alignment robustness. On three model families, alignment widens the gap on 97.5-99.8\% of toxic prompts, and gap closure tracks True ASR across suffix strategies (an internal consistency check, since our method optimises for gap closure). We present logit-gap steering, a gradient-free, forward-pass-only method that searches for in-distribution suffixes drawn from the model's own high-probability candidates whose cumulative effect closes the gap. Discovering 8 ensemble suffixes per family costs ${\approx}26{,}000$ forward-pass equivalents (${\approx}2$~min on one A100), ${\approx}125\times$ less than a single GCG search. Suffixes discovered on 0.5B-2B models transfer to 72B within family. An 8-suffix ensemble reaches 38--96\% True ASR across 13 models on AdvBench and HarmBench. Most suffixes have $10^{3}$-$10^{4}\times$ lower perplexity than GCG: under a published PPL-filter defense, our ensemble holds at 76.0\% (from 76.9\%) while GCG collapses from 64.7\% to 1.0\%.

The ability to reliably distinguish human-written text from that generated by large language models is of profound societal importance. The dominant approach to this problem exploits the likelihood hypothesis: that machine-generated text should appear more probable to a detector language model than human-written text. However, we demonstrate that the token-level signal distinguishing human and machine text is non-uniform across the hidden space of the detector model, and naively averaging likelihood-based token scores across regions with fundamentally different statistical structure, as most detectors do, causes a form of Simpson's paradox: a strong local signal is destroyed by inappropriate aggregation. To correct for this, we introduce a learned local calibration step grounded in Bayesian decision theory. Rather than aggregating raw token scores, we first learn lightweight predictors of the score distributions conditioned on position in hidden space, and aggregate calibrated log-likelihood ratios instead. This single intervention dramatically and consistently improves detection performance across all baseline detectors and all datasets we consider. For example, our calibrated variant of Fast-DetectGPT improves AUROC from $0.63$ to $0.85$ on GPT-5.4 text, and a locally-calibrated DMAP detector we introduce achieves state-of-the-art performance across the board. That said, our central contribution is not a new detector, but a precise diagnosis of a significant cause of under-performance of existing detectors and a principled, modular remedy compatible with any token-averaging pipeline. This will serve as a foundation for the community to build upon, with natural avenues including richer distributional models, improved calibration strategies, and principled ensembling with hidden-space geometry signals via the full Bayes-optimal decision rule.

Time-series data are often sampled irregularly at high frequency and exhibit long-range dependencies, which makes long-horizon modelling difficult. Continuous-time models such as neural controlled differential equations (NCDEs) and neural rough differential equations (NRDEs) can handle irregular sampling, but they scale poorly to long sequences. Selective state-space models (SSMs) such as Mamba scale linearly with sequence length, but within a single block they provide limited recurrent mixing across hidden dimensions. We propose LogSig-SSM (Log-Signature Compression for State-Space Models), which first compresses long multivariate time series into a shorter sequence of tokens using multi-scale windowed log-signatures, and then processes these tokens with a selective SSM backbone. LogSig-SSM is scalable and robust to irregular sampling, combining log-signature tokens that capture higher-order cross-channel interactions with a selective SSM that models long-range dependencies. The model also admits a continuous-time interpretation as an NCDE/NRDE-style system driven by a log-signature-based input, in which selectivity induces an input-dependent rescaling of the latent dynamics. Across four benchmarks, namely long-sequence classification on UEA, high-frequency physiological regression on PPG-DaLiA, multivariate weather forecasting, and irregularly sampled clinical prediction on PhysioNet Sepsis, LogSig-SSM overall outperforms existing SSM and continuous-time baselines while significantly reducing training time and GPU memory usage.

Vision-Language Models (VLMs) have achieved substantial progress across a wide range of understanding and reasoning tasks, driven by large-scale image-text training aimed at multimodal fusion. Ideally, replacing a textual question with its rendered-image counterpart should leave model performance essentially unaffected. In practice, however, such modality substitution induces dramatic performance degradation. We attribute this carrier sensitivity issue to an inherent bias in current training corpora. Across prevalent datasets such as image captioning, VQA, OCR, and web-sourced interleaved data, text and images are typically organized into distinct and asymmetric roles, with text serving as linguistic queries and images as visual references. Such data bias leads VLMs to exhibit distinct preferences for information acquisition across different modalities. Consequently, VLMs fail to align representations of semantically equivalent content across textual and visual carriers, making model reasoning fragile under modality substitution. To address this, we propose Local Modality Substitution (LoMo), a lightweight, architecture-agnostic data curation paradigm designed to provide supervision for cross-modal representational invariance between semantically equivalent text and image carriers. LoMo achieves this by reformulating single-modality prompts into seamlessly interleaved multimodal sequences. It dynamically selects target text spans and recasts them as rendered images, thereby preserving the same semantics across "text, visual, text" carriers. Extensive experiments across 13 diverse multimodal benchmarks demonstrate that LoMo significantly improves overall multimodal reasoning and yields deeper cross-modal fusion. Specifically, it delivers consistent gains across foundational models, improving over standard SFT by 2.67 points on LLaVA-OneVision-1.5-8B and 2.82 points on Qwen3.5-9B.

We introduce LongBanana, to our knowledge the \textbf{first expert-designed and expert-verified benchmark} for \emph{long-context multi-reference image synthesis} in the age of \textbf{agent-controlled image generation}. The task requires retrieving dispersed visual evidence, rejecting distractors, binding entities and relations, and composing one coherent final image from multiple references. LongBanana contains 90 carefully designed tasks organized by a 3-by-3 taxonomy; each task includes about 32 reference images, 20 constraints, and 50 evaluation items, yielding \textbf{4{,}670 expert-specified atomic evaluation items} in total. Its graph-grounded evaluation design instantiates each task with reference images and an entity-relation graph, then converts node, edge, and whole-scene obligations into checklist items for Content Fidelity, Relational Binding, and Global Constraint Satisfaction. Reviewed expert labels include a 15\% double-labeled subset with \textbf{89.4\% exact agreement} and \textbf{0.83 quadratic weighted kappa}. We further introduce LongBanana-Gen, an agentic controller framework for long-horizon image generation; experiments show that minimal agent harnesses improve over direct generation, while ReAct/Plan-Exec/Reflect yields controller-dependent gains by strengthening evidence grounding and repair. Together, LongBanana provides an auditable testbed for measuring dense visual evidence use and a shared target for developing stronger agent-controlled image-generation systems.

Long-horizon agency requires an AI system to preserve task state across hours or days of work, many tool calls, and repeated context resets. A common response is to scale a single resource: the model’s context window, together with inference-time reasoning. We argue that this response conflates long horizon with long context. Durable long-horizon state should not live only in the prompt; it should be externalized into the harness, the non-model infrastructure that manages tools, files, plans, checkpoints, sub-agents, permissions, and recovery. Publicly documented long-running agent systems already rely on such structures, including progress files, plan files, git history, rollback checkpoints, scratchpads, sub-agent decomposition, and browser-state checkpoints. We support this position with three lines of argument. First, memory-hierarchy history shows that pressure to scale one working resource repeatedly gave way to externally managed hierarchical state. Second, we identify five structural mismatches between long-context scaling and long-horizon agency: degraded effective context, lossy compaction, a single trust region, no transactional semantics, and no working-set decomposition. Third, we map long-horizon requirements to what context scaling provides and what harness-managed state must supply. We conclude by proposing eight harness state-management patterns and arguing that harness state should become a first-class research object for long-horizon agency.

Off-policy, value-based reinforcement learning methods such as Q-learning are appealing because they can learn from arbitrary experience, including data collected by older policies or other agents. In practice, however, bootstrapping makes long-horizon learning brittle: estimation errors at later states propagate backward through temporal-difference (TD) updates and can compound over time. We propose long-horizon Q-learning (LQL), which introduces a principled backstop against compounding error when learning the optimal action-value function. LQL builds on a prior optimality tightening observation: any realized action sequence lower-bounds what the optimal policy can achieve in expectation, so acting optimally earlier should not be worse than following the observed actions for several steps before switching to optimal behavior. Our contribution is to turn this inequality into a practical stabilization mechanism for Q-learning by using a hinge loss to penalize violations of these bounds. Importantly, LQL computes these penalties using network outputs already produced for the TD error, requiring no auxiliary networks and no additional forward passes relative to Q-learning. When combined with multiple state-of-the-art methods on a range of online and offline-to-online benchmarks, LQL consistently outperforms both 1-step TD and n-step TD learning at similar runtime.

Embodied Visual Tracking (EVT) aims to autonomously control an agent to follow a target in 3D space, which is critical for robotic navigation and human-robot interaction. However, robust long-term EVT in real-world scenarios faces two key bottlenecks: frequent tracking failures caused by occlusions and distractors, and the lack of a long-term benchmark for model training and evaluation. To address these, we propose a systematic solution. First, we develop LT-VLA, a lightweight VLA model featuring a Target Memory module to maintain consistent target identification and an Adaptive Execution module to adjust tracking actions based on observation reliability. Second, we construct LTEVT, a unified long-term EVT benchmark comprising 11 high-fidelity indoor and outdoor environments, enriched with over 5,000 assets and 325 human targets to simulate occlusions and distractors. Along with the benchmark, we release the LTEVT-4500K dataset for large-scale training and a comprehensive evaluation toolkit. Extensive experiments demonstrate that our 0.6B model achieves state-of-the-art performance comparable to 7B-scale models while maintaining superior efficiency. Notably, LT-VLA achieves 62.4% average Success Rate on EVT-Bench and runs at 20 FPS on a Unitree G1 robot, delivering a practical and robust solution for real-world deployment.

Algorithmic risk predictors are increasingly being used to inform resource allocation in healthcare and other high-stakes domains. Common in many such settings, the public decision maker delegates service provision to private organizations and compensates them according to each individual's predicted cost of service. We show that this common allocation mechanism can fail when organizations engage in \emph{favorable selection}, i.e., strategically choosing individuals whose actual costs are below the payment model predictions. Motivated by health insurance payments in the Medicare Advantage (MA) program, we provide a theoretical framework to study the interaction between a decision maker and a strategically selecting service provider. Our theoretical framework predicts two empirical patterns observed in Medicare Advantage that have become central concerns: enrollment in the private program expands, while government payments rise above the counterfactual cost of direct public provision. Guided by our theoretical framework, we then propose a reformed payment policy that is implementable under the government's data-access constraints. We prove that its equilibria eliminate overpayment while preserving profits for efficient organizations. We validate the framework in synthetic and semi-synthetic experiments that illustrate the advantages of our reformed policy, over the status quo.

Generative policies have emerged as a compelling paradigm in reinforcement learning (RL) as they can represent complex and multimodal action distributions. However, this high expressivity leaves them particularly vulnerable to the inherent target distribution non-stationarity in RL, rendering the generative process severely volatile. To address this issue, we propose \textbf{L}ook-ah\textbf{E}ad v\textbf{A}riational \textbf{F}low (LEAF), a generative policy optimization framework that stabilizes both the target and the generative process. First, we replace the action-value-guided target with a look-ahead target induced by predictive dynamics, which we theoretically prove to exhibit a tighter target drift bound. Second, we formulate policy improvement as a proximal variational free-energy optimization, which enforces a distributional trust region to prevent overfitting to transient target shifts. Finally, we prove that this optimization induces an optimal transport flow, which we instantiate as a two-phase coarse-to-fine generative process to achieve broad multimodal coverage and pinpoint accuracy. Extensive experiments on the DeepMind Control Suite and Humanoid Bench demonstrate that LEAF consistently outperforms strong baselines in both sample efficiency and asymptotic performance.

Transformers have achieved remarkable success across a wide range of applications, and a growing body of work suggests that part of their strength comes from their ability to learn and execute algorithmic procedures. However, our understanding of how transformers learn such algorithms remains limited, especially in the presence of layer normalization (LN). In this work, we study principal component prediction as a concrete testbed for understanding the training dynamics of transformers with LN. We prove that a looped linear transformer with LN, trained by gradient descent, converges to a solution that implements the power method, with each self-attention layer performing one power iteration. Notably, the model is trained only for principal component prediction, rather than being explicitly supervised to implement the power method. Our finding thus reveals an ``algorithmic implicit bias'' of looped transformers with LN: principal-component prediction can in principle be achieved by many mechanisms, yet gradient descent selects one that realizes the power method. We further provide a concrete comparison between transformers with and without LN: even with layerwise guidance from power iterations, a transformer without LN cannot exactly learn the power method, whereas the corresponding transformer with LN can, leading to a provable performance gap in principal component prediction. Our results provide, to our knowledge, the first theoretical analysis of the training dynamics of looped and single-layer transformers with LN, and shed light on the role of LN in transformer models.

Inverse Reinforcement Learning (IRL) aims to recover a reward function that explains expert demonstrations. Existing IRL methods typically rely on a bi-level optimization procedure that alternates between reward learning and policy optimization, leading to substantial computational burden and training instability. In this work, we introduce a different route that eliminates policy optimization entirely by leveraging diffusion policies. Our key insight is that a diffusion policy encodes the action-gradient structure of the optimal soft Q function, enabling reward learning to be cast as a sequence of value recovery problems, thereby allowing us to bypass reward-policy loops inherent in prior IRL methods. Specifically, our method proceeds in three stages: (I) recovering the optimal soft Q function via action-gradient matching and estimating the corresponding soft value function (LogSumExp of Q values) in a way inspired by Gumbel regression; (II) calibrating these soft values by inferring a state-dependent offset; (III) extracting the reward by enforcing Bellman consistency. This leads to Loop-Free Inverse Reinforcement Learning (LFIRL), a fully offline algorithm that operates in a simple, loop-free, and sequential manner. LFIRL is simple to implement, significantly improves training efficiency while maintaining strong reward recovery performance. Empirically, across Maze, Franka Kitchen, Adroit Hand Pen, and Push-T benchmarks, LFIRL achieves 2-3x speedup over the fastest baselines, while matching or surpassing state-of-the-art methods in reward recovery quality.