Sketch-fusion: A gradient compression method with multi-layer fusion for communication-efficient distributed training

doi:10.1016/j.jpdc.2023.104811

CSpace

	Sketch-fusion: A gradient compression method with multi-layer fusion for communication-efficient distributed training
	Dai, Lingfei 1,2; Gong, Luqi 1; An, Zhulin 1; Xu, Yongjun 1; Diao, Boyu 1
	2024-03-01
发表期刊	JOURNAL OF PARALLEL AND DISTRIBUTED COMPUTING
ISSN	0743-7315
卷号	185 页码:10
摘要	Gradient compression is an effective technique for improving the efficiency of distributed training. However, introducing gradient compression can reduce model accuracy and training efficiency. Furthermore, we also find that using a layer-wise gradient compression algorithm would lead to significant compression and communication overhead, which can negatively impact the scaling efficiency of the distributed training system. To address these issues, we propose a new method called Sketch-Fusion SGD, which leverages the Count-Sketch data structure to enhance the scalability and training speed of distributed deep learning systems. Moreover, our method employs LayerFusion to optimize gradient compression algorithms' scalability and convergence efficiency by formulating an optimal multi-layer fusion strategy without introducing extra hyperparameters. We evaluate our method on a cluster of 16 GPUs and demonstrate that it can improve training efficiency by up to 18.6% without compromising the model's accuracy. In addition, we find that applying our LayerFusion algorithm to other gradient compression methods improved their scalability by up to 2.87x.
关键词	Gradient compression Multi-layer fusion Distributed stochastic gradient descent Deep learning training
DOI	10.1016/j.jpdc.2023.104811
收录类别	SCI
语种	英语
WOS研究方向	Computer Science
WOS类目	Computer Science, Theory & Methods
WOS记录号	WOS:001127654600001
出版者	ACADEMIC PRESS INC ELSEVIER SCIENCE
引用统计	被引频次：1[WOS] [WOS记录] [WOS相关记录]
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/38454
专题	中国科学院计算技术研究所
通讯作者	Diao, Boyu
作者单位	1.Chinese Acad Sci, Inst Comp Technol, Beijing, Peoples R China 2.Univ Chinese Acad Sci, Coll Comp Sci, Beijing, Peoples R China
推荐引用方式 GB/T 7714	Dai, Lingfei,Gong, Luqi,An, Zhulin,et al. Sketch-fusion: A gradient compression method with multi-layer fusion for communication-efficient distributed training[J]. JOURNAL OF PARALLEL AND DISTRIBUTED COMPUTING,2024,185:10.
APA	Dai, Lingfei,Gong, Luqi,An, Zhulin,Xu, Yongjun,&Diao, Boyu.(2024).Sketch-fusion: A gradient compression method with multi-layer fusion for communication-efficient distributed training.JOURNAL OF PARALLEL AND DISTRIBUTED COMPUTING,185,10.
MLA	Dai, Lingfei,et al."Sketch-fusion: A gradient compression method with multi-layer fusion for communication-efficient distributed training".JOURNAL OF PARALLEL AND DISTRIBUTED COMPUTING 185(2024):10.