Kenneth Tran

It looks like there's a bug in the [benchmark of Chainer](https://github.com/soumith/convnet-benchmarks/blob/master/chainer/train_imagenet.py#L98). They computed the averaged time to be `total / niter-1` instead of `total / (niter-1)`.  --- Another thing...

Hi @huawei-bigdata , Will do when I have the time to use them and study in more depth. If someone has experience with one of these packages, (s)he certainly can...

Thanks @sirotenko 1. Agreed that distributed training of CNNs is harder. However, keep in mind that: (a) although there aren't many empirical evidences, 1-bit (or 2/4-bit) quantization of the gradient...