Decentralized distributed deep learning in heterogeneous WAN environments

Rankyung Hong, Abhishek Chandra

Research output: Chapter in Book/Report/Conference proceedingConference contribution

2 Scopus citations
Original languageEnglish (US)
Title of host publicationSoCC 2018 - Proceedings of the 2018 ACM Symposium on Cloud Computing
PublisherAssociation for Computing Machinery, Inc
Pages505
Number of pages1
ISBN (Electronic)9781450360111
DOIs
StatePublished - Oct 11 2018
Event2018 ACM Symposium on Cloud Computing, SoCC 2018 - Carlsbad, United States
Duration: Oct 11 2018Oct 13 2018

Publication series

NameSoCC 2018 - Proceedings of the 2018 ACM Symposium on Cloud Computing

Other

Other2018 ACM Symposium on Cloud Computing, SoCC 2018
CountryUnited States
CityCarlsbad
Period10/11/1810/13/18

Bibliographical note

Funding Information:
However, the size of parameters shared every iteration is fixed across machines in the existing work since they do not consider heterogeneous networks. Moreover the data size trained in unit time can vary depending on computing power. Thus, it is important that the frequency, size of parameters as well as its contribution to parameter updates should be more precisely and dynamically controlled by the DL framework based on different individual system resources and environments for better model accuracy. Acknowledgement. This work is supported in part by NSF grant III-1422802.

Funding Information:
This work is supported in part by NSF grant III-1422802.

Copyright:
Copyright 2018 Elsevier B.V., All rights reserved.

Cite this