MARINA: Faster Non-Convex Distributed Learning with Compression
Eduard Gorbunov 1 2 3 Konstantin Burlachenko 3 Zhize Li 3 Peter Richtarik 3
Abstract ment of new efficient methods for non-convex optimization
We develop and analyze MARINA: a new com- problems (Danilova et al., 2020), and especially the training
munication efficient method for non-convex dis- of deep learning models (Sun, 2019).
tributed learning over heterogeneous d ...


雷达卡




京公网安备 11010802022788号







