File Download

There are no files associated with this item.

  Links for fulltext
     (May Require Subscription)
Supplementary

Conference Paper: Stochastic composite mirror descent: Optimal bounds with high probabilities

TitleStochastic composite mirror descent: Optimal bounds with high probabilities
Authors
Issue Date2018
Citation
Advances in Neural Information Processing Systems, 2018, v. 2018-December, p. 1519-1529 How to Cite?
AbstractWe study stochastic composite mirror descent, a class of scalable algorithms able to exploit the geometry and composite structure of a problem. We consider both convex and strongly convex objectives with non-smooth loss functions, for each of which we establish high-probability convergence rates optimal up to a logarithmic factor. We apply the derived computational error bounds to study the generalization performance of multi-pass stochastic gradient descent (SGD) in a non-parametric setting. Our high-probability generalization bounds enjoy a loga-rithmical dependency on the number of passes provided that the step size sequence is square-summable, which improves the existing bounds in expectation with a polynomial dependency and therefore gives a strong justification on the ability of multi-pass SGD to overcome overfitting. Our analysis removes boundedness assumptions on subgradients often imposed in the literature. Numerical results are reported to support our theoretical findings.
Persistent Identifierhttp://hdl.handle.net/10722/329561
ISSN
2020 SCImago Journal Rankings: 1.399

 

DC FieldValueLanguage
dc.contributor.authorLei, Yunwen-
dc.contributor.authorTang, Ke-
dc.date.accessioned2023-08-09T03:33:41Z-
dc.date.available2023-08-09T03:33:41Z-
dc.date.issued2018-
dc.identifier.citationAdvances in Neural Information Processing Systems, 2018, v. 2018-December, p. 1519-1529-
dc.identifier.issn1049-5258-
dc.identifier.urihttp://hdl.handle.net/10722/329561-
dc.description.abstractWe study stochastic composite mirror descent, a class of scalable algorithms able to exploit the geometry and composite structure of a problem. We consider both convex and strongly convex objectives with non-smooth loss functions, for each of which we establish high-probability convergence rates optimal up to a logarithmic factor. We apply the derived computational error bounds to study the generalization performance of multi-pass stochastic gradient descent (SGD) in a non-parametric setting. Our high-probability generalization bounds enjoy a loga-rithmical dependency on the number of passes provided that the step size sequence is square-summable, which improves the existing bounds in expectation with a polynomial dependency and therefore gives a strong justification on the ability of multi-pass SGD to overcome overfitting. Our analysis removes boundedness assumptions on subgradients often imposed in the literature. Numerical results are reported to support our theoretical findings.-
dc.languageeng-
dc.relation.ispartofAdvances in Neural Information Processing Systems-
dc.titleStochastic composite mirror descent: Optimal bounds with high probabilities-
dc.typeConference_Paper-
dc.description.naturelink_to_subscribed_fulltext-
dc.identifier.scopuseid_2-s2.0-85064828146-
dc.identifier.volume2018-December-
dc.identifier.spage1519-
dc.identifier.epage1529-

Export via OAI-PMH Interface in XML Formats


OR


Export to Other Non-XML Formats