Robust Training in High Dimensions via Block Coordinate Geometric Median Descent

Anish Acharya; A Hashemi; P Jain; S Sanghavi; I Dhillon; U Topcu

Robust Training in High Dimensions via Block Coordinate Geometric Median Descent

Anish Acharya, A Hashemi, P Jain, S Sanghavi, I Dhillon, U Topcu

International Conference on Artificial Intelligence and Statistics (AISTATS), 2022 · Spotlight — recorded talk, virtual conference

Makes geometric-median SGD practical in high dimensions by applying it to one chosen block of coordinates at a time, keeping the optimal 1/2 breakdown point for smooth non-convex problems.

PMLR Code Poster Slides

Short version: Joint IFML/CCSI Symposium (Simons Institute, UC Berkeley) · Short version: NSF-TRIPODS Workshop on Communication Efficient Distributed Optimization

All publications · Research