Show simple item record

dc.contributor.authorSapio, Amedeo
dc.contributor.authorCanini, Marco
dc.contributor.authorHo, Chen-Yu
dc.contributor.authorNelson, Jacob
dc.contributor.authorKalnis, Panos
dc.contributor.authorKim, Changhoon
dc.contributor.authorKrishnamurthy, Arvind
dc.contributor.authorMoshref, Masoud
dc.contributor.authorPorts, Dan R. K.
dc.contributor.authorRichtarik, Peter
dc.date.accessioned2019-05-28T13:11:04Z
dc.date.available2019-05-28T13:11:04Z
dc.date.issued2019-02-22
dc.identifier.urihttp://hdl.handle.net/10754/653105
dc.description.abstractTraining complex machine learning models in parallel is an increasinglyimportant workload. We accelerate distributed parallel training by designing acommunication primitive that uses a programmable switch dataplane to execute akey step of the training process. Our approach, SwitchML, reduces the volume ofexchanged data by aggregating the model updates from multiple workers in thenetwork. We co-design the switch processing with the end-host protocols and MLframeworks to provide a robust, efficient solution that speeds up training byup to 300%, and at least by 20% for a number of real-world benchmark models.
dc.publisherarXiv
dc.relation.urlhttps://arxiv.org/abs/1903.06701
dc.relation.urlhttps://arxiv.org/pdf/1903.06701
dc.rightsArchived with thanks to arXiv
dc.titleScaling Distributed Machine Learning with In-Network Aggregation
dc.typePreprint
dc.contributor.departmentComputer Science
dc.contributor.departmentComputer Science Program
dc.contributor.departmentComputer, Electrical and Mathematical Sciences and Engineering (CEMSE) Division
dc.contributor.departmentExtreme Computing Research Center
dc.eprint.versionPre-print
dc.contributor.institutionMicrosoft
dc.contributor.institutionBarefoot Networks
dc.contributor.institutionUniversity of Washington
dc.identifier.arxivid1903.06701
kaust.personSapio, Amedeo
kaust.personCanini, Marco
kaust.personHo, Chen-Yu
kaust.personKalnis, Panos
kaust.personRichtarik, Peter
dc.versionv1
refterms.dateFOA2019-05-28T13:11:19Z
display.summary<p>This record has been merged with an existing record at: <a href="http://hdl.handle.net/10754/631179">http://hdl.handle.net/10754/631179</a>.</p>


Files in this item

Thumbnail
Name:
1903.06701.pdf
Size:
995.6Kb
Format:
PDF
Description:
Preprint

This item appears in the following Collection(s)

Show simple item record