Revisiting Distributed Synchronous SGD.

The recent success of deep learning approaches for domains like speech recognition (Hinton et al., 2012) and computer vision (Ioffe & Szegedy, 2015) stems from many algorithmic improvements but also from the fact that the size of available training data has grown significantly over the years, together with the computing power, in terms of both CPUs and GPUs. While a single GPU often provides algorithmic simplicity and speed up to a given scale of data and model, there exist an operating point where a distributed implementation of training algorithms for deep architectures becomes necessary. Previous works have been focusing on asynchronous SGD training, which works well up to a few dozens of workers for some models. In this work, we show that synchronous SGD training, with the help of backup workers, can not only achieve better accuracy, but also reach convergence faster with respect to wall time, i.e. use more workers more efficiently.