BriefGPT.xyz
Apr, 2019
大步长同步分布式 SGD 的通信权衡
Communication trade-offs for synchronized distributed SGD with large step size
HTML
PDF
Kumar Kshitij Patel, Aymeric Dieuleveut
TL;DR
本文提出了一种名为local-SGD的算法,通过逐步同步而非每一步都进行通信提高了通信效率,同时在大步长情况下提供了自适应下限比较。
Abstract
Synchronous
mini-batch sgd
is state-of-the-art for large-scale
distributed machine learning
. However, in practice, its
convergence
is bott
→