We develop simple differentially private optimization algorithms that move along directions of (expected) descent to find an approximate second-order solution for nonconvex ERM. We use line search, mini-batching, and a two-phase strategy to improve the speed and practicality of the algorithm. Numerical experiments demonstrate the effectiveness of these approaches.
@article{arxiv.2302.04972,
title = {Differentially Private Optimization for Smooth Nonconvex ERM},
author = {Changyu Gao and Stephen J. Wright},
journal= {arXiv preprint arXiv:2302.04972},
year = {2023}
}