Det-CGD: Compressed Gradient Descent with Matrix Stepsizes for Non-Convex Optimization.

Hanmin Li, Avetik G. Karagulyan, Peter Richtrik

Browse the full ICLR paper archive.