Y. Lin, S. Han, H. Mao, Y. Wang, и W. Dally. (2017)cite arxiv:1712.01887Comment: we find 99.9% of the gradient exchange in distributed SGD is redundant; we reduce the communication bandwidth by two orders of magnitude without losing accuracy.
T. Wang, X. Zhou, W. Zhang, и J. Wei. Proceedings of the Second Asia-Pacific Symposium on Internetware, стр. 18:1--18:4. New York, NY, USA, ACM, (2010)