Robust Learning for Data Poisoning Attacks

Yunjuan Wang, Poorya Mianjy, Raman Arora
Proceedings of the 38th International Conference on Machine Learning, PMLR 139:10859-10869, 2021.

Abstract

We investigate the robustness of stochastic approximation approaches against data poisoning attacks. We focus on two-layer neural networks with ReLU activation and show that under a specific notion of separability in the RKHS induced by the infinite-width network, training (finite-width) networks with stochastic gradient descent is robust against data poisoning attacks. Interestingly, we find that in addition to a lower bound on the width of the network, which is standard in the literature, we also require a distribution-dependent upper bound on the width for robust generalization. We provide extensive empirical evaluations that support and validate our theoretical results.

Cite this Paper


BibTeX
@InProceedings{pmlr-v139-wang21r, title = {Robust Learning for Data Poisoning Attacks}, author = {Wang, Yunjuan and Mianjy, Poorya and Arora, Raman}, booktitle = {Proceedings of the 38th International Conference on Machine Learning}, pages = {10859--10869}, year = {2021}, editor = {Meila, Marina and Zhang, Tong}, volume = {139}, series = {Proceedings of Machine Learning Research}, month = {18--24 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v139/wang21r/wang21r.pdf}, url = {https://proceedings.mlr.press/v139/wang21r.html}, abstract = {We investigate the robustness of stochastic approximation approaches against data poisoning attacks. We focus on two-layer neural networks with ReLU activation and show that under a specific notion of separability in the RKHS induced by the infinite-width network, training (finite-width) networks with stochastic gradient descent is robust against data poisoning attacks. Interestingly, we find that in addition to a lower bound on the width of the network, which is standard in the literature, we also require a distribution-dependent upper bound on the width for robust generalization. We provide extensive empirical evaluations that support and validate our theoretical results.} }
Endnote
%0 Conference Paper %T Robust Learning for Data Poisoning Attacks %A Yunjuan Wang %A Poorya Mianjy %A Raman Arora %B Proceedings of the 38th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2021 %E Marina Meila %E Tong Zhang %F pmlr-v139-wang21r %I PMLR %P 10859--10869 %U https://proceedings.mlr.press/v139/wang21r.html %V 139 %X We investigate the robustness of stochastic approximation approaches against data poisoning attacks. We focus on two-layer neural networks with ReLU activation and show that under a specific notion of separability in the RKHS induced by the infinite-width network, training (finite-width) networks with stochastic gradient descent is robust against data poisoning attacks. Interestingly, we find that in addition to a lower bound on the width of the network, which is standard in the literature, we also require a distribution-dependent upper bound on the width for robust generalization. We provide extensive empirical evaluations that support and validate our theoretical results.
APA
Wang, Y., Mianjy, P. & Arora, R.. (2021). Robust Learning for Data Poisoning Attacks. Proceedings of the 38th International Conference on Machine Learning, in Proceedings of Machine Learning Research 139:10859-10869 Available from https://proceedings.mlr.press/v139/wang21r.html.

Related Material