MARINA: Faster Non-Convex Distributed Learning with Compression
Abstract
We develop and analyze MARINA: a new communication efficient method for non-convex distributed learning over heterogeneous datasets. MARINA employs a novel communication compression strategy based on the compression of gradient differences that is reminiscent of but different from the strategy employed in the DIANA method of Mishchenko et al. (2019). Unlike virtually all competing distributed first-order methods, including DIANA, ours is based on a carefully designed biased gradient estimator, which is the key to its superior theoretical and practical performance. The communication complexity bounds we prove for MARINA are evidently better than those of all previous first-order methods. Further, we develop and analyze two variants of MARINA: VR-MARINA and PP-MARINA. The first method is designed for the case when the local loss functions owned by clients are either of a finite sum or of an expectation form, and the second method allows for a partial participation of clients -- a feature important in federated learning. All our methods are superior to previous state-of-the-art methods in terms of oracle/communication complexity. Finally, we provide a convergence analysis of all methods for problems satisfying the Polyak-Lojasiewicz condition.
Cite
@article{arxiv.2102.07845,
title = {MARINA: Faster Non-Convex Distributed Learning with Compression},
author = {Eduard Gorbunov and Konstantin Burlachenko and Zhize Li and Peter Richtárik},
journal= {arXiv preprint arXiv:2102.07845},
year = {2022}
}
Comments
ICML 2021; v2 contains additional experiments; v3 contains small fixes of the typos in the complexity results for online case for VR-MARINA; 42 pages, 6 figures, 3 algorithms