DC-S3GD: Delay-Compensated Stale-Synchronous SGD for Large-Scale Decentralized Neural Network Training
This addresses the challenge of scalable and efficient neural network training for large-scale distributed systems, though it appears incremental as it builds on existing delay-compensated methods.
The paper tackled the problem of training deep neural networks efficiently in a decentralized setting by proposing DC-S3GD, a delay-compensated stale-synchronous SGD algorithm, which achieved state-of-the-art results with large batches.
Data parallelism has become the de facto standard for training Deep Neural Network on multiple processing units. In this work we propose DC-S3GD, a decentralized (without Parameter Server) stale-synchronous version of the Delay-Compensated Asynchronous Stochastic Gradient Descent (DC-ASGD) algorithm. In our approach, we allow for the overlap of computation and communication, and compensate the inherent error with a first-order correction of the gradients. We prove the effectiveness of our approach by training Convolutional Neural Network with large batches and achieving state-of-the-art results.