Gossip training for deep learning

Michael Blot, David Picard, Matthieu Cord +1

We address the issue of speeding up the training of convolutional networks. Here we study a distributed method adapted to stochastic gradient descent (SGD). The parallel optimizati…