Reviews: Pipe-SGD: A Decentralized Pipelined SGD Framework for Distributed Deep Net Training

This paper proposed a pipelined training setup for neural nets. The pipeline structure allows computation and communication to be carried out concurrently within each worker. Experiments show speed improvements over existing methods. Overall I think the method and the evaluations are convincing. However, I believe the NIPS conference is not the appropriate venue for this paper for the following reasons.