Exploiting Structure in Crowdsourcing Tasks via Latent Factor Models 基于潜在因素模型的众包任务开发结构

Exploiting Structure in Crowdsourcing Tasks via Latent Factor Models 基于潜在因素模型的众包任务开发结构

ID:18435916

大小:450.25 KB

页数:9页

时间:2018-09-17

Exploiting Structure in Crowdsourcing Tasks via Latent Factor Models 基于潜在因素模型的众包任务开发结构_第1页
Exploiting Structure in Crowdsourcing Tasks via Latent Factor Models 基于潜在因素模型的众包任务开发结构_第2页
Exploiting Structure in Crowdsourcing Tasks via Latent Factor Models 基于潜在因素模型的众包任务开发结构_第3页
Exploiting Structure in Crowdsourcing Tasks via Latent Factor Models 基于潜在因素模型的众包任务开发结构_第4页
Exploiting Structure in Crowdsourcing Tasks via Latent Factor Models 基于潜在因素模型的众包任务开发结构_第5页
资源描述:

《Exploiting Structure in Crowdsourcing Tasks via Latent Factor Models 基于潜在因素模型的众包任务开发结构》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库

1、ExploitingStructureinCrowdsourcingTasksviaLatentFactorModelsPaulRuvolo,JacobWhitehill,JavierR.MovellanMachinePerceptionLaboratoryUCSanDiego,CAAbstractInternetcrowdsourcingservicessuchastheAmazonMechanicalTurk(1)andtheESPGame(15)havebecomeimportanttoolsforthemachinelear

2、ningcommu-nitybyfacilitatingthedistributedlabelingoflargedatasetsatlittlecost.Akeychallengewhenusingcrowdsourcingtolabeldatabasesistheneedtoderivehighqualitylabelsbyaggregatingtheresponsesfromlabelersofvaryingreliabilityoverdatainstancesofvaryingdifficulty.Existingalgor

3、ithmsforqualitycontrolandlabelinference(14;17;10)sufferseveralsignificantshortcomings:(1)Exist-ingmethodsareincapableofmodelinginteractioneffectsbetweenlabeleranddataitems,suchaswhensomelabelershavespecializedknowledgeaboutaparticularsubsetofitems.(2)Existingalgorithmsa

4、ssumethatlabelers’accuracies,aswellasdatainstances’difficulties,areindependent.Inreality,theremaybeaprioriin-formationaboutlabelers(ordatainstances)thatpredictsthoselabelers’accuracyatthelabelingtask.Analogously,certainfeaturessharedamongdatainstancesmaypredicttheirdiffi

5、cultyofbeinglabeledcorrectly.Inthispaper,wepresentanalgorithmthataddressesbothoftheseshortcomings.Wedemonstratethattheproposedalgorithmdeliverssuperioraccuracy,comparedtopreviousmethods,ofinferringdatalabelsonadifficultfacialexpressionlabelingtask.Finally,weshowthatourp

6、roposedmodelsubsumescertainpreviousmodelsasspecialcases.1IntroductionAsmachinelearningapplicationstacklemoreandmoredifficultproblems,theimportanceoflarge-scaleandvarieddatasetsofhigh-qualitytrainingdatabecomesmoreapparent.Forinstance,theOmronfacedetector,whichrepresents

7、thecurrentstateoftheartoffacedetection,wastrainedusingmillionsofhand-labeledimages(9).Inordertomeetthelargedemandforlabeleddata,researchershavebeenincreasinglyrelyingoncrowdsourcingservicesthatallowthemtoharnessvastpoolsofhumanlabelersatverylowcost.Currentsystemsinclud

8、ePay-per-labelservicessuchasAmazon’sMechanicalTurk(1)andinteractivegamessuchasHerdIt(2)andtheESPgame(15).Ascrowdsourc

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。