Timezone: »
Stochastic Gradient Descent (SGD) has been the method of choice for learning large-scale non-convex models. While a general analysis of when SGD works has been elusive, there has been a lot of recent progress in understanding the convergence of Gradient Flow (GF) on the population loss, partly due to the simplicity that a continuous-time analysis buys us. An overarching theme of our paper is providing general conditions under which SGD converges, assuming that GF on the population loss converges. Our main tool to establish this connection is a general \textit{converse Lyapunov} like theorem, which implies the existence of a Lyapunov potential under mild assumptions on the rates of convergence of GF. In fact, using these potentials, we show a one-to-one correspondence between rates of convergence of GF and geometrical properties of the underlying objective. When these potentials further satisfy certain self-bounding properties, we show that they can be used to provide a convergence guarantee for Gradient Descent (GD) and SGD (even when the GF path and GD/SGD paths are quite far apart). It turns out that these self-bounding assumptions are in a sense also necessary for GD/SGD to work. Using our framework, we provide a unified analysis for GD/SGD not only for classical settings like convex losses, or objectives that satisfy PL/ KL properties, but also for more complex problems including Phase Retrieval and Matrix sq-root, and extending the results in the recent work of Chatterjee 2022.
Author Information
Christopher De Sa (Cornell University)
Satyen Kale (Google)
Jason Lee (University of Southern California)
Ayush Sekhari (Cornell University)
Karthik Sridharan (Cornell University)
More from the Same Authors
-
2022 : Self-Stabilization: The Implicit Bias of Gradient Descent at the Edge of Stability »
Alex Damian · Eshaan Nichani · Jason Lee -
2022 : Hybrid RL: Using both offline and online data can make RL efficient »
Yuda Song · Yifei Zhou · Ayush Sekhari · J. Bagnell · Akshay Krishnamurthy · Wen Sun -
2022 : Hidden Poison: Machine Unlearning Enables Camouflaged Poisoning Attacks »
Jimmy Di · Jack Douglas · Jayadev Acharya · Gautam Kamath · Ayush Sekhari -
2022 : Hidden Poison: Machine unlearning enables camouflaged poisoning attacks »
Jimmy Di · Jack Douglas · Jayadev Acharya · Gautam Kamath · Ayush Sekhari -
2022 Panel: Panel 4A-2: Adaptively Exploiting d-Separators… & On the Complexity… »
Blair Bilodeau · Ayush Sekhari -
2022 Poster: Identifying good directions to escape the NTK regime and efficiently learn low-degree plus sparse polynomials »
Eshaan Nichani · Yu Bai · Jason Lee -
2022 Poster: Model Preserving Compression for Neural Networks »
Jerry Chee · Megan Flynn (née Renz) · Anil Damle · Christopher De Sa -
2022 Poster: Provably Efficient Reinforcement Learning in Partially Observable Dynamical Systems »
Masatoshi Uehara · Ayush Sekhari · Jason Lee · Nathan Kallus · Wen Sun -
2022 Poster: Implicit Bias of Gradient Descent on Reparametrized Models: On Equivalence to Mirror Descent »
Zhiyuan Li · Tianhao Wang · Jason Lee · Sanjeev Arora -
2022 Poster: Understanding Hyperdimensional Computing for Parallel Single-Pass Learning »
Tao Yu · Yichi Zhang · Zhiru Zhang · Christopher De Sa -
2022 Poster: GraB: Finding Provably Better Data Permutations than Random Reshuffling »
Yucheng Lu · Wentao Guo · Christopher De Sa -
2022 Poster: Reproducibility in Optimization: Theoretical Framework and Limits »
Kwangjun Ahn · Prateek Jain · Ziwei Ji · Satyen Kale · Praneeth Netrapalli · Gil I Shamir -
2022 Poster: On the Effective Number of Linear Regions in Shallow Univariate ReLU Networks: Convergence Guarantees and Implicit Bias »
Itay Safran · Gal Vardi · Jason Lee -
2022 Poster: On the Complexity of Adversarial Decision Making »
Dylan J Foster · Alexander Rakhlin · Ayush Sekhari · Karthik Sridharan -
2021 Poster: How Fine-Tuning Allows for Effective Meta-Learning »
Kurtland Chua · Qi Lei · Jason Lee -
2021 Poster: Label Noise SGD Provably Prefers Flat Global Minimizers »
Alex Damian · Tengyu Ma · Jason Lee -
2021 Poster: Going Beyond Linear RL: Sample Efficient Neural Function Approximation »
Baihe Huang · Kaixuan Huang · Sham Kakade · Jason Lee · Qi Lei · Runzhe Wang · Jiaqi Yang -
2021 Poster: Predicting What You Already Know Helps: Provable Self-Supervised Learning »
Jason Lee · Qi Lei · Nikunj Saunshi · JIACHENG ZHUO -
2021 Poster: Optimal Gradient-based Algorithms for Non-concave Bandit Optimization »
Baihe Huang · Kaixuan Huang · Sham Kakade · Jason Lee · Qi Lei · Runzhe Wang · Jiaqi Yang -
2020 Poster: Online learning with dynamics: A minimax perspective »
Kush Bhatia · Karthik Sridharan -
2020 Poster: Reinforcement Learning with Feedback Graphs »
Christoph Dann · Yishay Mansour · Mehryar Mohri · Ayush Sekhari · Karthik Sridharan -
2019 Poster: Hypothesis Set Stability and Generalization »
Dylan Foster · Spencer Greenberg · Satyen Kale · Haipeng Luo · Mehryar Mohri · Karthik Sridharan -
2018 : Contributed Talk 1 »
Jason Lee -
2018 Poster: Implicit Bias of Gradient Descent on Linear Convolutional Networks »
Suriya Gunasekar · Jason Lee · Daniel Soudry · Nati Srebro -
2018 Poster: Algorithmic Regularization in Learning Deep Homogeneous Models: Layers are Automatically Balanced »
Simon Du · Wei Hu · Jason Lee -
2018 Poster: Adding One Neuron Can Eliminate All Bad Local Minima »
SHIYU LIANG · Ruoyu Sun · Jason Lee · R. Srikant -
2018 Poster: Uniform Convergence of Gradients for Non-Convex Learning and Optimization »
Dylan Foster · Ayush Sekhari · Karthik Sridharan -
2018 Poster: Provably Correct Automatic Sub-Differentiation for Qualified Programs »
Sham Kakade · Jason Lee -
2018 Poster: On the Convergence and Robustness of Training GANs with Regularized Optimal Transport »
Maziar Sanjabi · Jimmy Ba · Meisam Razaviyayn · Jason Lee -
2017 Poster: Gradient Descent Can Take Exponential Time to Escape Saddle Points »
Simon Du · Chi Jin · Jason D Lee · Michael Jordan · Aarti Singh · Barnabas Poczos -
2017 Poster: Parameter-Free Online Learning via Model Selection »
Dylan J Foster · Satyen Kale · Mehryar Mohri · Karthik Sridharan -
2017 Spotlight: Parameter-Free Online Learning via Model Selection »
Dylan J Foster · Satyen Kale · Mehryar Mohri · Karthik Sridharan -
2017 Spotlight: Gradient Descent Can Take Exponential Time to Escape Saddle Points »
Simon Du · Chi Jin · Jason D Lee · Michael Jordan · Aarti Singh · Barnabas Poczos -
2016 Poster: Exploiting the Structure: Stochastic Gradient Methods Using Raw Clusters »
Zeyuan Allen-Zhu · Yang Yuan · Karthik Sridharan -
2016 Poster: Learning in Games: Robustness of Fast Convergence »
Dylan Foster · zhiyuan li · Thodoris Lykouris · Karthik Sridharan · Eva Tardos -
2016 Oral: Matrix Completion has No Spurious Local Minimum »
Rong Ge · Jason Lee · Tengyu Ma -
2016 Poster: Matrix Completion has No Spurious Local Minimum »
Rong Ge · Jason Lee · Tengyu Ma -
2015 Poster: Adaptive Online Learning »
Dylan Foster · Alexander Rakhlin · Karthik Sridharan -
2015 Spotlight: Adaptive Online Learning »
Dylan Foster · Alexander Rakhlin · Karthik Sridharan -
2015 Poster: Evaluating the statistical significance of biclusters »
Jason D Lee · Yuekai Sun · Jonathan E Taylor -
2014 Poster: Scalable Methods for Nonnegative Matrix Factorizations of Near-separable Tall-and-skinny Matrices »
Austin Benson · Jason D Lee · Bartek Rajwa · David F Gleich -
2014 Spotlight: Scalable Methods for Nonnegative Matrix Factorizations of Near-separable Tall-and-skinny Matrices »
Austin Benson · Jason D Lee · Bartek Rajwa · David F Gleich -
2014 Poster: Exact Post Model Selection Inference for Marginal Screening »
Jason D Lee · Jonathan E Taylor -
2013 Poster: On model selection consistency of penalized M-estimators: a geometric theory »
Jason D Lee · Yuekai Sun · Jonathan E Taylor -
2013 Poster: Using multiple samples to learn mixture models »
Jason D Lee · Ran Gilad-Bachrach · Rich Caruana -
2013 Spotlight: Using multiple samples to learn mixture models »
Jason D Lee · Ran Gilad-Bachrach · Rich Caruana -
2012 Poster: Proximal Newton-type Methods for Minimizing Convex Objective Functions in Composite Form »
Jason D Lee · Yuekai Sun · Michael Saunders -
2010 Poster: Practical Large-Scale Optimization for Max-norm Regularization »
Jason D Lee · Benjamin Recht · Russ Salakhutdinov · Nati Srebro · Joel A Tropp