Abstract: Gradient compression algorithms are widely used to alleviate the communication bottleneck in distributed ML. However, existing gradient compression algorithms suffer from accuracy ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results