Adversarial Training for Gradient Descent: Analysis Through its Continuous-time Approximation
Machine Learning
2023-05-25 v2 Optimization and Control
Probability
Abstract
Adversarial training has gained great popularity as one of the most effective defenses for deep neural network and more generally for gradient-based machine learning models against adversarial perturbations on data points. This paper establishes a continuous-time approximation for the mini-max game of adversarial training. This approximation approach allows for precise and analytical comparisons between stochastic gradient descent and its adversarial training counterpart; and confirms theoretically the robustness of adversarial training from a new gradient-flow viewpoint. The analysis is then corroborated through various analytical and numerical examples.
Cite
@article{arxiv.2105.08037,
title = {Adversarial Training for Gradient Descent: Analysis Through its Continuous-time Approximation},
author = {Haotian Gu and Xin Guo and Xinyu Li},
journal= {arXiv preprint arXiv:2105.08037},
year = {2023}
}