Revisiting Adversarial Risk

Arun Sai Suggala, Adarsh Prasad, Vaishnavh Nagarajan, Pradeep Ravikumar
Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, PMLR 89:2331-2339, 2019.

Abstract

Recent works on adversarial perturbations show that there is an inherent trade-off between standard test accuracy and adversarial accuracy. Specifically, they show that no classifier can simultaneously be robust to adversarial perturbations and achieve high standard test accuracy. However, this is contrary to the standard notion that on tasks such as image classification, humans are robust classifiers with low error rate. In this work, we show that the main reason behind this confusion is the inaccurate definition of adversarial perturbation that is used in the literature. To fix this issue, we propose a slight, yet important modification to the existing definition of adversarial perturbation. Based on the modified definition, we show that there is no trade-off between adversarial and standard accuracies; there exist classifiers that are robust and achieve high standard accuracy. We further study several properties of this new definition of adversarial risk and its relation to the existing definition.

Cite this Paper


BibTeX
@InProceedings{pmlr-v89-suggala19a, title = {Revisiting Adversarial Risk}, author = {Suggala, Arun Sai and Prasad, Adarsh and Nagarajan, Vaishnavh and Ravikumar, Pradeep}, booktitle = {Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics}, pages = {2331--2339}, year = {2019}, editor = {Chaudhuri, Kamalika and Sugiyama, Masashi}, volume = {89}, series = {Proceedings of Machine Learning Research}, month = {16--18 Apr}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v89/suggala19a/suggala19a.pdf}, url = {https://proceedings.mlr.press/v89/suggala19a.html}, abstract = {Recent works on adversarial perturbations show that there is an inherent trade-off between standard test accuracy and adversarial accuracy. Specifically, they show that no classifier can simultaneously be robust to adversarial perturbations and achieve high standard test accuracy. However, this is contrary to the standard notion that on tasks such as image classification, humans are robust classifiers with low error rate. In this work, we show that the main reason behind this confusion is the inaccurate definition of adversarial perturbation that is used in the literature. To fix this issue, we propose a slight, yet important modification to the existing definition of adversarial perturbation. Based on the modified definition, we show that there is no trade-off between adversarial and standard accuracies; there exist classifiers that are robust and achieve high standard accuracy. We further study several properties of this new definition of adversarial risk and its relation to the existing definition.} }
Endnote
%0 Conference Paper %T Revisiting Adversarial Risk %A Arun Sai Suggala %A Adarsh Prasad %A Vaishnavh Nagarajan %A Pradeep Ravikumar %B Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2019 %E Kamalika Chaudhuri %E Masashi Sugiyama %F pmlr-v89-suggala19a %I PMLR %P 2331--2339 %U https://proceedings.mlr.press/v89/suggala19a.html %V 89 %X Recent works on adversarial perturbations show that there is an inherent trade-off between standard test accuracy and adversarial accuracy. Specifically, they show that no classifier can simultaneously be robust to adversarial perturbations and achieve high standard test accuracy. However, this is contrary to the standard notion that on tasks such as image classification, humans are robust classifiers with low error rate. In this work, we show that the main reason behind this confusion is the inaccurate definition of adversarial perturbation that is used in the literature. To fix this issue, we propose a slight, yet important modification to the existing definition of adversarial perturbation. Based on the modified definition, we show that there is no trade-off between adversarial and standard accuracies; there exist classifiers that are robust and achieve high standard accuracy. We further study several properties of this new definition of adversarial risk and its relation to the existing definition.
APA
Suggala, A.S., Prasad, A., Nagarajan, V. & Ravikumar, P.. (2019). Revisiting Adversarial Risk. Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 89:2331-2339 Available from https://proceedings.mlr.press/v89/suggala19a.html.

Related Material