Timezone: »
Post-hoc gradient-based interpretability methods [Simonyan et al., 2013, Smilkov et al., 2017] that provide instance-specific explanations of model predictions are often based on assumption (A): magnitude of input gradients—gradients of logits with respect to input—noisily highlight discriminative task-relevant features. In this work, we test the validity of assumption (A) using a three-pronged approach:1. We develop an evaluation framework, DiffROAR, to test assumption (A) on four image classification benchmarks. Our results suggest that (i) input gradients of standard models (i.e., trained on original data) may grossly violate (A), whereas (ii) input gradients of adversarially robust models satisfy (A).2. We then introduce BlockMNIST, an MNIST-based semi-real dataset, that by design encodes a priori knowledge of discriminative features. Our analysis on BlockMNIST leverages this information to validate as well as characterize differences between input gradient attributions of standard and robust models.3. Finally, we theoretically prove that our empirical findings hold on a simplified version of the BlockMNIST dataset. Specifically, we prove that input gradients of standard one-hidden-layer MLPs trained on this dataset do not highlight instance-specific signal coordinates, thus grossly violating assumption (A).Our findings motivate the need to formalize and test common assumptions in interpretability in a falsifiable manner [Leavitt and Morcos, 2020]. We believe that the DiffROAR evaluation framework and BlockMNIST-based datasets can serve as sanity checks to audit instance-specific interpretability methods; code and data available at https://github.com/harshays/inputgradients.
Author Information
Harshay Shah (Google Research)
Prateek Jain (Google Research)
Praneeth Netrapalli (Google)
More from the Same Authors
-
2021 Spotlight: Near-optimal Offline and Streaming Algorithms for Learning Non-Linear Dynamical Systems »
Suhas Kowshik · Dheeraj Nagaraj · Prateek Jain · Praneeth Netrapalli -
2021 Spotlight: Differentially Private Model Personalization »
Prateek Jain · John Rush · Adam Smith · Shuang Song · Abhradeep Guha Thakurta -
2021 Spotlight: Near-Optimal Lower Bounds For Convex Optimization For All Orders of Smoothness »
Ankit Garg · Robin Kothari · Praneeth Netrapalli · Suhail Sherif -
2022 : MET: Masked Encoding for Tabular Data »
Kushal Majmundar · Sachin Goyal · Praneeth Netrapalli · Prateek Jain -
2022 : A Unified Framework for Comparing Learning Algorithms »
Harshay Shah · Sung Min Park · Andrew Ilyas · Aleksander Madry -
2022 : Learning an Invertible Output Mapping Can Mitigate Simplicity Bias in Neural Networks »
Sravanti Addepalli · Anshul Nasery · Venkatesh Babu R · Praneeth Netrapalli · Prateek Jain -
2023 Poster: AdANNS: A Framework for Adaptive Semantic Search »
Aniket Rege · Aditya Kusupati · Sharan Ranjit S · Alan Fan · Qingqing Cao · Sham Kakade · Prateek Jain · Ali Farhadi -
2023 Poster: Simplicity Bias in 1-Hidden Layer Neural Networks »
Depen Morwani · Jatin Batra · Prateek Jain · Praneeth Netrapalli -
2023 Poster: Blocked Collaborative Bandits: Online Collaborative Filtering with Per-Item Budget Constraints »
Soumyabrata Pal · Arun Suggala · Karthikeyan Shanmugam · Prateek Jain -
2023 Poster: Label Robust and Differentially Private Linear Regression: Computational and Statistical Efficiency »
Xiyang Liu · Prateek Jain · Weihao Kong · Sewoong Oh · Arun Suggala -
2022 Poster: DP-PCA: Statistically Optimal and Differentially Private PCA »
Xiyang Liu · Weihao Kong · Prateek Jain · Sewoong Oh -
2022 Poster: S3GC: Scalable Self-Supervised Graph Clustering »
Fnu Devvrit · Aditya Sinha · Inderjit Dhillon · Prateek Jain -
2022 Poster: Reproducibility in Optimization: Theoretical Framework and Limits »
Kwangjun Ahn · Prateek Jain · Ziwei Ji · Satyen Kale · Praneeth Netrapalli · Gil I Shamir -
2022 Poster: Matryoshka Representation Learning »
Aditya Kusupati · Gantavya Bhatt · Aniket Rege · Matthew Wallingford · Aditya Sinha · Vivek Ramanujan · William Howard-Snyder · Kaifeng Chen · Sham Kakade · Prateek Jain · Ali Farhadi -
2021 Poster: Differentially Private Model Personalization »
Prateek Jain · John Rush · Adam Smith · Shuang Song · Abhradeep Guha Thakurta -
2021 Poster: Streaming Linear System Identification with Reverse Experience Replay »
Suhas Kowshik · Dheeraj Nagaraj · Prateek Jain · Praneeth Netrapalli -
2021 Poster: LLC: Accurate, Multi-purpose Learnt Low-dimensional Binary Codes »
Aditya Kusupati · Matthew Wallingford · Vivek Ramanujan · Raghav Somani · Jae Sung Park · Krishna Pillutla · Prateek Jain · Sham Kakade · Ali Farhadi -
2021 Poster: Near-optimal Offline and Streaming Algorithms for Learning Non-Linear Dynamical Systems »
Suhas Kowshik · Dheeraj Nagaraj · Prateek Jain · Praneeth Netrapalli -
2021 Poster: Statistically and Computationally Efficient Linear Meta-representation Learning »
Kiran Thekumparampil · Prateek Jain · Praneeth Netrapalli · Sewoong Oh -
2021 Poster: Near-Optimal Lower Bounds For Convex Optimization For All Orders of Smoothness »
Ankit Garg · Robin Kothari · Praneeth Netrapalli · Suhail Sherif -
2020 Poster: Projection Efficient Subgradient Method and Optimal Nonsmooth Frank-Wolfe Method »
Kiran Thekumparampil · Prateek Jain · Praneeth Netrapalli · Sewoong Oh -
2020 Spotlight: Projection Efficient Subgradient Method and Optimal Nonsmooth Frank-Wolfe Method »
Kiran Thekumparampil · Prateek Jain · Praneeth Netrapalli · Sewoong Oh -
2020 Poster: RNNPool: Efficient Non-linear Pooling for RAM Constrained Inference »
Oindrila Saha · Aditya Kusupati · Harsha Vardhan Simhadri · Manik Varma · Prateek Jain -
2020 Spotlight: RNNPool: Efficient Non-linear Pooling for RAM Constrained Inference »
Oindrila Saha · Aditya Kusupati · Harsha Vardhan Simhadri · Manik Varma · Prateek Jain -
2020 Poster: The Pitfalls of Simplicity Bias in Neural Networks »
Harshay Shah · Kaustav Tamuly · Aditi Raghunathan · Prateek Jain · Praneeth Netrapalli -
2020 Poster: Least Squares Regression with Markovian Data: Fundamental Limits and Algorithms »
Dheeraj Nagaraj · Xian Wu · Guy Bresler · Prateek Jain · Praneeth Netrapalli -
2020 Spotlight: Least Squares Regression with Markovian Data: Fundamental Limits and Algorithms »
Dheeraj Nagaraj · Xian Wu · Guy Bresler · Prateek Jain · Praneeth Netrapalli