Robust Learning for Data Poisoning Attacks

Yunjuan Wang · Poorya Mianjy · Raman Arora

[ Abstract ] [ Livestream: Visit Adversarial Learning 3 ] [ Paper ]
Thu 22 Jul 5:40 p.m. — 5:45 p.m. PDT
[ Paper ]

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.

Chat is not available.