arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2004.05298cs.LGcs.CVstat.ML
收藏

分布式随机稀疏化SGD的分离误差反馈

Detached Error Feedback for Distributed SGD with Random Sparsification

University of Pittsburgh(匹兹堡大学)

An Xu, Heng Huang

摘要:

通信瓶颈一直是大规模分布式深度学习中的关键问题。在这项工作中,我们研究了分布式SGD与随机块稀疏作为梯度压缩机,这是环allreduce兼容和高计算效率,但导致性能较差。为了解决这一重要问题,我们从一个新的方面,即方差和梯度的二阶矩之间的权衡,提高了通信效率的分布式SGD。基于这种动机,我们提出了一种新的分离式误差反馈(DEF)算法,该算法在非凸问题上显示出比误差反馈更好的收敛界。我们还提出了DEF-A在训练的早期阶段加速DEF的泛化,这表明了比DEF更好的泛化范围。此外,我们首次建立了通信高效的分布式SGD和具有平均值的SGD(SGD-IA)之间的联系。广泛的深度学习实验表明,在各种设置下,所提出的方法有显着的经验改进。

英文摘要:

The communication bottleneck has been a critical problem in large-scale distributed deep learning. In this work, we study distributed SGD with random block-wise sparsification as the gradient compressor, which is ring-allreduce compatible and highly computation-efficient but leads to inferior performance. To tackle this important issue, we improve the communication-efficient distributed SGD from a novel aspect, that is, the trade-off between the variance and second moment of the gradient. With this motivation, we propose a new detached error feedback (DEF) algorithm, which shows better convergence bound than error feedback for non-convex problems. We also propose DEF-A to accelerate the generalization of DEF at the early stages of the training, which shows better generalization bounds than DEF. Furthermore, we establish the connection between communication-efficient distributed SGD and SGD with iterate averaging (SGD-IA) for the first time. Extensive deep learning experiments show significant empirical improvement of the proposed methods under various settings.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑