Shu Hu

Differentially Private SGDA for Minimax Problems

2022 · arXiv (Cornell University)

Stochastic gradient descent ascent (SGDA) and its variants have been the workhorse for solving minimax problems. However, in contrast to the well-studied stochastic gradient descent (SGD) with differential privacy (DP) constraints, there is little work …

Outlier Robust Adversarial Training

2023 · arXiv (Cornell University)

Supervised learning models are challenged by the intrinsic complexities of training data such as outliers and minority subpopulations and intentional attacks at inference time with adversarial samples. While traditional robust learning methods and the recent …