[1903.10635] Federated Learning Of Out-Of-Vocabulary Words

文章推薦指數: 80 %
投票人數:10人

We demonstrate that a character-level recurrent neural network is able to learn out-of-vocabulary (OOV) words under federated learning settings, ... GlobalSurvey Injust3minutes,helpusbetterunderstandhowyouperceivearXiv. Takethesurvey TAKESURVEY ComputerScience>ComputationandLanguage arXiv:1903.10635(cs) [Submittedon26Mar2019] Title:FederatedLearningOfOut-Of-VocabularyWords Authors:MingqingChen,RajivMathews,TomOuyang,FrançoiseBeaufays DownloadPDF Abstract:Wedemonstratethatacharacter-levelrecurrentneuralnetworkisableto learnout-of-vocabulary(OOV)wordsunderfederatedlearningsettings,forthe purposeofexpandingthevocabularyofavirtualkeyboardforsmartphones withoutexportingsensitivetexttoservers.High-frequencywordscanbe sampledfromthetrainedgenerativemodelbydrawingfromthejointposterior directly.Westudythefeasibilityoftheapproachintwosettings:(1)using simulatedfederatedlearningonapubliclyavailablenon-IIDper-userdataset fromapopularsocialnetworkingwebsite,(2)usingfederatedlearningondata hostedonusermobiledevices.Themodelachievesgoodrecallandprecision comparedtoground-truthOOVwordsinsetting(1).With(2)wedemonstratethe practicalityofthisapproachbyshowingthatwecanlearnmeaningfulOOVwords withgoodcharacter-levelpredictionaccuracyandcrossentropyloss. Subjects: ComputationandLanguage(cs.CL) Citeas: arXiv:1903.10635[cs.CL]   (or arXiv:1903.10635v1[cs.CL]forthisversion)   https://doi.org/10.48550/arXiv.1903.10635 Focustolearnmore arXiv-issuedDOIviaDataCite SubmissionhistoryFrom:MingqingChen[viewemail] [v1] Tue,26Mar201900:01:47UTC(765KB) Full-textlinks: Download: PDF Otherformats (license) Currentbrowsecontext:cs.CL new | recent | 1903 Changetobrowseby: cs References&Citations NASAADSGoogleScholar SemanticScholar DBLP-CSBibliography listing|bibtex MingqingChenRajivMathewsTomOuyangFrançoiseBeaufays a exportbibtexcitation Loading... Bibtexformattedcitation × loading... Dataprovidedby: Bookmark BibliographicTools BibliographicandCitationTools BibliographicExplorerToggle BibliographicExplorer(WhatistheExplorer?) LitmapsToggle Litmaps(WhatisLitmaps?) scite.aiToggle sciteSmartCitations(WhatareSmartCitations?) Code&Data CodeandDataAssociatedwiththisArticle arXivLinkstoCodeToggle arXivLinkstoCode&Data(WhatisLinkstoCode&Data?) Demos Demos ReplicateToggle Replicate(WhatisReplicate?) RelatedPapers RecommendersandSearchTools ConnectedPapersToggle ConnectedPapers(WhatisConnectedPapers?) Corerecommendertoggle CORERecommender(WhatisCORE?) AboutarXivLabs arXivLabs:experimentalprojectswithcommunitycollaborators arXivLabsisaframeworkthatallowscollaboratorstodevelopandsharenewarXivfeaturesdirectlyonourwebsite. BothindividualsandorganizationsthatworkwitharXivLabshaveembracedandacceptedourvaluesofopenness,community,excellence,anduserdataprivacy.arXiviscommittedtothesevaluesandonlyworkswithpartnersthatadheretothem. HaveanideaforaprojectthatwilladdvalueforarXiv'scommunity?LearnmoreaboutarXivLabsandhowtogetinvolved. Whichauthorsofthispaperareendorsers?| DisableMathJax(WhatisMathJax?)



請為這篇文章評分?