TernGrad: Ternary Gradients to Reduce Communication in Distributed Deep Learning

Open in new window