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 ...