Journal
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS
Volume 32, Issue 7, Pages 1725-1739Publisher
IEEE COMPUTER SOC
DOI: 10.1109/TPDS.2020.3040606
Keywords
Computational modeling; Training; Convergence; Program processors; Stochastic processes; Deep learning; Task analysis; Stochastic gradient descent; distributed deep learning; decentralized optimization
Funding
- European Research Council (ERC) under the European Union [678880, 801039]
- ERC [805223]
- Swiss National Science Foundation [185778]
- ETH Postdoctoral Fellowship
- European Research Council (ERC) [805223] Funding Source: European Research Council (ERC)
Ask authors/readers for more resources
The study introduces a wait-avoiding stochastic optimizer, WAGMA-SGD, that reduces global communication through subgroup weight exchange while maintaining convergence rates similar to globally communicating SGD. Empirical results show significant advantages of the method across different tasks, particularly in training throughput and time-to-solution.
Deep learning at scale is dominated by communication time. Distributing samples across nodes usually yields the best performance, but poses scaling challenges due to global information dissemination and load imbalance across uneven sample lengths. State-of-the-art decentralized optimizers mitigate the problem, but require more iterations to achieve the same accuracy as their globally-communicating counterparts. We present Wait-Avoiding Group Model Averaging (WAGMA) SGD, a wait-avoiding stochastic optimizer that reduces global communication via subgroup weight exchange. The key insight is a combination of algorithmic changes to the averaging scheme and the use of a group allreduce operation. We prove the convergence of WAGMA-SGD, and empirically show that it retains convergence rates similar to Allreduce-SGD. For evaluation, we train ResNet-50 on ImageNet; Transformer for machine translation; and deep reinforcement learning for navigation at scale. Compared with state-of-the-art decentralized SGD variants, WAGMA-SGD significantly improves training throughput (e.g., 2.1x on 1,024 GPUs for reinforcement learning), and achieves the fastest time-to-solution (e.g., the highest score using the shortest training time for Transformer).
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available