PowerGossip: Practical Low-Rank Communication Compression in Decentralized Deep Learning
This addresses the communication overhead problem for decentralized training over arbitrary networks, offering a practical solution with incremental improvements in simplicity and convergence speed.
The paper tackles the communication bottleneck in decentralized deep learning by introducing PowerGossip, a simple algorithm that compresses model differences using low-rank linear compressors with power iteration, requiring no additional hyperparameters and converging faster than prior methods, performing on par with state-of-the-art tuned compression algorithms in benchmarks.
Lossy gradient compression has become a practical tool to overcome the communication bottleneck in centrally coordinated distributed training of machine learning models. However, algorithms for decentralized training with compressed communication over arbitrary connected networks have been more complicated, requiring additional memory and hyperparameters. We introduce a simple algorithm that directly compresses the model differences between neighboring workers using low-rank linear compressors applied on model differences. Inspired by the PowerSGD algorithm for centralized deep learning, this algorithm uses power iteration steps to maximize the information transferred per bit. We prove that our method requires no additional hyperparameters, converges faster than prior methods, and is asymptotically independent of both the network and the compression. Out of the box, these compressors perform on par with state-of-the-art tuned compression algorithms in a series of deep learning benchmarks.