Vector quantization for distributed machine learning
We aim at reducing the communication cost between workers in a distributed machine learning system. Our approach is to compress/quantize the gradients generated in each learning itertaion. Our goal is to find linear and unbiased compressor.