Do Perceptually Aligned Gradients Imply Robustness?

Roy Ganz, Bahjat Kawar, Michael Elad
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:10628-10648, 2023.

Abstract

Adversarially robust classifiers possess a trait that non-robust models do not - Perceptually Aligned Gradients (PAG). Their gradients with respect to the input align well with human perception. Several works have identified PAG as a byproduct of robust training, but none have considered it as a standalone phenomenon nor studied its own implications. In this work, we focus on this trait and test whether Perceptually Aligned Gradients imply Robustness. To this end, we develop a novel objective to directly promote PAG in training classifiers and examine whether models with such gradients are more robust to adversarial attacks. Extensive experiments on multiple datasets and architectures validate that models with aligned gradients exhibit significant robustness, exposing the surprising bidirectional connection between PAG and robustness. Lastly, we show that better gradient alignment leads to increased robustness and harness this observation to boost the robustness of existing adversarial training techniques.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-ganz23a, title = {Do Perceptually Aligned Gradients Imply Robustness?}, author = {Ganz, Roy and Kawar, Bahjat and Elad, Michael}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {10628--10648}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/ganz23a/ganz23a.pdf}, url = {https://proceedings.mlr.press/v202/ganz23a.html}, abstract = {Adversarially robust classifiers possess a trait that non-robust models do not - Perceptually Aligned Gradients (PAG). Their gradients with respect to the input align well with human perception. Several works have identified PAG as a byproduct of robust training, but none have considered it as a standalone phenomenon nor studied its own implications. In this work, we focus on this trait and test whether Perceptually Aligned Gradients imply Robustness. To this end, we develop a novel objective to directly promote PAG in training classifiers and examine whether models with such gradients are more robust to adversarial attacks. Extensive experiments on multiple datasets and architectures validate that models with aligned gradients exhibit significant robustness, exposing the surprising bidirectional connection between PAG and robustness. Lastly, we show that better gradient alignment leads to increased robustness and harness this observation to boost the robustness of existing adversarial training techniques.} }
Endnote
%0 Conference Paper %T Do Perceptually Aligned Gradients Imply Robustness? %A Roy Ganz %A Bahjat Kawar %A Michael Elad %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-ganz23a %I PMLR %P 10628--10648 %U https://proceedings.mlr.press/v202/ganz23a.html %V 202 %X Adversarially robust classifiers possess a trait that non-robust models do not - Perceptually Aligned Gradients (PAG). Their gradients with respect to the input align well with human perception. Several works have identified PAG as a byproduct of robust training, but none have considered it as a standalone phenomenon nor studied its own implications. In this work, we focus on this trait and test whether Perceptually Aligned Gradients imply Robustness. To this end, we develop a novel objective to directly promote PAG in training classifiers and examine whether models with such gradients are more robust to adversarial attacks. Extensive experiments on multiple datasets and architectures validate that models with aligned gradients exhibit significant robustness, exposing the surprising bidirectional connection between PAG and robustness. Lastly, we show that better gradient alignment leads to increased robustness and harness this observation to boost the robustness of existing adversarial training techniques.
APA
Ganz, R., Kawar, B. & Elad, M.. (2023). Do Perceptually Aligned Gradients Imply Robustness?. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:10628-10648 Available from https://proceedings.mlr.press/v202/ganz23a.html.

Related Material