Federated Optimization Algorithms with Random Reshuffling and Gradient Compression
From MaRDI portal
Publication:6402104
arXiv2206.07021MaRDI QIDQ6402104
Author name not available (Why is that?)
Publication date: 14 June 2022
Abstract: Gradient compression is a popular technique for improving communication complexity of stochastic first-order methods in distributed training of machine learning models. However, the existing works consider only with-replacement sampling of stochastic gradients. In contrast, it is well-known in practice and recently confirmed in theory that stochastic methods based on without-replacement sampling, e.g., Random Reshuffling (RR) method, perform better than ones that sample the gradients with-replacement. In this work, we close this gap in the literature and provide the first analysis of methods with gradient compression and without-replacement sampling. We first develop a na"ive combination of random reshuffling with gradient compression (Q-RR). Perhaps surprisingly, but the theoretical analysis of Q-RR does not show any benefits of using RR. Our extensive numerical experiments confirm this phenomenon. This happens due to the additional compression variance. To reveal the true advantages of RR in the distributed learning with compression, we propose a new method called DIANA-RR that reduces the compression variance and has provably better convergence rates than existing counterparts with with-replacement sampling of stochastic gradients. Next, to have a better fit to Federated Learning applications, we incorporate local computation, i.e., we propose and analyze the variants of Q-RR and DIANA-RR -- Q-NASTYA and DIANA-NASTYA that use local gradient steps and different local and global stepsizes. Finally, we conducted several numerical experiments to illustrate our theoretical results.
Has companion code repository: https://github.com/igorsokoloff/rr_with_compression_experiments_source_code
This page was built for publication: Federated Optimization Algorithms with Random Reshuffling and Gradient Compression
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6402104)
