Log In Sign Up

Provable Robustness of Adversarial Training for Learning Halfspaces with Noise

by   Difan Zou, et al.

We analyze the properties of adversarial training for learning adversarially robust halfspaces in the presence of agnostic label noise. Denoting 𝖮𝖯𝖳_p,r as the best robust classification error achieved by a halfspace that is robust to perturbations of ℓ_p balls of radius r, we show that adversarial training on the standard binary cross-entropy loss yields adversarially robust halfspaces up to (robust) classification error Õ(√(𝖮𝖯𝖳_2,r)) for p=2, and Õ(d^1/4√(𝖮𝖯𝖳_∞, r) + d^1/2𝖮𝖯𝖳_∞,r) when p=∞. Our results hold for distributions satisfying anti-concentration properties enjoyed by log-concave isotropic distributions among others. We additionally show that if one instead uses a nonconvex sigmoidal loss, adversarial training yields halfspaces with an improved robust classification error of O(𝖮𝖯𝖳_2,r) for p=2, and O(d^1/4𝖮𝖯𝖳_∞, r) when p=∞. To the best of our knowledge, this is the first work to show that adversarial training provably yields robust classifiers in the presence of noise.


page 1

page 2

page 3

page 4


Fooling Adversarial Training with Inducing Noise

Adversarial training is widely believed to be a reliable approach to imp...

On Adversarial Risk and Training

In this work we formally define the notions of adversarial perturbations...

Double Descent in Adversarial Training: An Implicit Label Noise Perspective

Here, we show that the robust overfitting shall be viewed as the early p...

MACER: Attack-free and Scalable Robust Training via Maximizing Certified Radius

Adversarial training is one of the most popular ways to learn robust mod...

The Consistency of Adversarial Training for Binary Classification

Robustness to adversarial perturbations is of paramount concern in moder...

Over-parameterized Adversarial Training: An Analysis Overcoming the Curse of Dimensionality

Adversarial training is a popular method to give neural nets robustness ...

Bayesian Inference with Certifiable Adversarial Robustness

We consider adversarial training of deep neural networks through the len...