Quantized Decentralized Stochastic Learning over Directed Graphs
This addresses communication efficiency for distributed machine learning systems, but it is incremental as it builds on existing push-sum algorithms.
The paper tackles the heavy communication bottleneck in decentralized stochastic learning over directed graphs by proposing a quantized algorithm, achieving the same convergence rates as exact-communication methods for convex and non-convex losses, with numerical evaluations showing significant speed-up.
We consider a decentralized stochastic learning problem where data points are distributed among computing nodes communicating over a directed graph. As the model size gets large, decentralized learning faces a major bottleneck that is the heavy communication load due to each node transmitting large messages (model updates) to its neighbors. To tackle this bottleneck, we propose the quantized decentralized stochastic learning algorithm over directed graphs that is based on the push-sum algorithm in decentralized consensus optimization. More importantly, we prove that our algorithm achieves the same convergence rates of the decentralized stochastic learning algorithm with exact-communication for both convex and non-convex losses. Numerical evaluations corroborate our main theoretical results and illustrate significant speed-up compared to the exact-communication methods.