dc.contributor.author Horvath, Samuel dc.contributor.author Richtarik, Peter dc.date.accessioned 2020-06-29T07:50:16Z dc.date.available 2020-06-29T07:50:16Z dc.date.issued 2020-06-19 dc.identifier.uri http://hdl.handle.net/10754/663911 dc.description.abstract Modern large-scale machine learning applications require stochastic optimization algorithms to be implemented on distributed compute systems. A key bottleneck of such systems is the communication overhead for exchanging information across the workers, such as stochastic gradients. Among the many techniques proposed to remedy this issue, one of the most successful is the framework of compressed communication with error feedback (EF). EF remains the only known technique that can deal with the error induced by contractive compressors which are not unbiased, such as Top-$K$. In this paper, we propose a new and theoretically and practically better alternative to EF for dealing with contractive compressors. In particular, we propose a construction which can transform any contractive compressor into an induced unbiased compressor. Following this transformation, existing methods able to work with unbiased compressors can be applied. We show that our approach leads to vast improvements over EF, including reduced memory requirements, better communication complexity guarantees and fewer assumptions. We further extend our results to federated learning with partial participation following an arbitrary distribution over the nodes, and demonstrate the benefits thereof. We perform several numerical experiments which validate our theoretical findings. dc.publisher arXiv dc.relation.url https://arxiv.org/pdf/2006.11077 dc.rights Archived with thanks to arXiv dc.title A Better Alternative to Error Feedback for Communication-Efficient Distributed Learning dc.type Preprint dc.contributor.department Statistics Program dc.contributor.department Statistics dc.contributor.department Computer Science Program dc.contributor.department Computer, Electrical and Mathematical Sciences and Engineering (CEMSE) Division dc.eprint.version Pre-print dc.identifier.arxivid 2006.11077 kaust.person Horvath, Samuel kaust.person Richtarik, Peter refterms.dateFOA 2020-06-29T07:51:05Z
﻿

### Files in this item

Name:
Preprintfile1.pdf
Size:
4.168Mb
Format:
PDF
Description:
Pre-print