Consensus Control for Decentralized Deep Learning
This work addresses the problem of performance degradation in decentralized deep learning for researchers and practitioners, offering theoretical insights and practical guidelines to mitigate this issue.
This paper investigates the performance degradation in decentralized deep learning, identifying the changing consensus distance between devices as a key explanatory parameter. They theoretically demonstrate that decentralized training converges as fast as centralized training when the consensus distance is below a critical threshold, and empirically validate this relationship.
Decentralized training of deep learning models enables on-device learning over networks, as well as efficient scaling to large compute clusters. Experiments in earlier works reveal that, even in a data-center setup, decentralized training often suffers from the degradation in the quality of the model: the training and test performance of models trained in a decentralized fashion is in general worse than that of models trained in a centralized fashion, and this performance drop is impacted by parameters such as network size, communication topology and data partitioning. We identify the changing consensus distance between devices as a key parameter to explain the gap between centralized and decentralized training. We show in theory that when the training consensus distance is lower than a critical quantity, decentralized training converges as fast as the centralized counterpart. We empirically validate that the relation between generalization performance and consensus distance is consistent with this theoretical observation. Our empirical insights allow the principled design of better decentralized training schemes that mitigate the performance drop. To this end, we provide practical training guidelines and exemplify its effectiveness on the data-center setup as the important first step.