Sparse Communication for Distributed Gradient Descent
Preprint
Abstract
We make distributed stochastic gradient descent faster by exchanging sparse updates instead of dense updates. Gradient updates are positively skewed as most updates are near zero, so we map the 99% smallest updates (by absolute value) to zero then exchange sparse matrices. This method can be combined with quantization to further improve the compression. We explore different configurations and apply them to neural machine translation and MNIST image classification tasks. Most configurations work on MNIST, whereas different configurations reduce convergence rate on the more complex translation task. Our experiments show that we can achieve up to 49% speed up on MNIST and 22% on NMT without damaging the final accuracy or BLEU.Keywords
All Related Versions
This publication has 4 references indexed in Scilit:
- Scalable distributed DNN training using commodity GPU cloud computing2015
- 1-bit stochastic gradient descent and its application to data-parallel distributed training of speech DNNs2014
- Fast k-selection algorithms for graphics processing unitsACM Journal of Experimental Algorithmics, 2012
- Gradient-based learning applied to document recognitionProceedings of the IEEE, 1998