Tuning Stochastic Gradient Algorithms for Statistical Inference via Large-Sample Asymptotics
Tuning Stochastic Gradient Algorithms for Statistical Inference via Large-Sample Asymptotics
About this item
Full title
Author / Creator
Publisher
Ithaca: Cornell University Library, arXiv.org
Journal title
Language
English
Formats
Publication information
Publisher
Ithaca: Cornell University Library, arXiv.org
Subjects
More information
Scope and Contents
Contents
The tuning of stochastic gradient algorithms (SGAs) for optimization and sampling is often based on heuristics and trial-and-error rather than generalizable theory. We address this theory--practice gap by characterizing the large-sample statistical asymptotics of SGAs via a joint step-size--sample-size scaling limit. We show that iterate averaging...
Alternative Titles
Full title
Tuning Stochastic Gradient Algorithms for Statistical Inference via Large-Sample Asymptotics
Authors, Artists and Contributors
Author / Creator
Identifiers
Primary Identifiers
Record Identifier
TN_cdi_proquest_journals_2694703048
Permalink
https://devfeature-collection.sl.nsw.gov.au/record/TN_cdi_proquest_journals_2694703048
Other Identifiers
E-ISSN
2331-8422
