英语汇报文稿-噪声标签-长尾学习-When Noisy Labels Meet Long Tail Dilemmas A Representation Calibration Method.pptx.docx

英语汇报文稿-噪声标签-长尾学习-When Noisy Labels Meet Long Tail Dilemmas A Representation Calibration Method.pptx.docx

  1. 1、本文档共2页,可阅读全部内容。
  2. 2、有哪些信誉好的足球投注网站(book118)网站文档一经付费(服务费),不意味着购买了该文档的版权,仅供个人/单位学习、研究之用,不得用于商业用途,未经授权,严禁复制、发行、汇编、翻译或者网络传播等,侵权必究。
  3. 3、本站所有内容均由合作方或网友上传,本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺!文档内容仅供研究参考,付费前请自行鉴别。如您付费,意味着您自己接受本站规则且自行承担风险,本站不退款、不进行额外附加服务;查看《如何避免下载的几个坑》。如果您已付费下载过本站文档,您可以点击 这里二次下载
  4. 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等,请点击“版权申诉”(推荐),也可以打举报电话:400-050-0827(电话支持时间:9:00-18:30)。
查看更多

SpeechTitle:PresentationforWhenNoisyLabelsMeetLongTailDilemmas:ARepresentationCalibrationMethod

姓名:?XXX????????班级:?专业学位英语A+10班???????????学号:SZ2316

正文:

Helloeveryone,Today,Iwouldliketoshareapaperwithyoutitled,WhenNoisyLabelsMeetLongTailDilemmas:ARepresentationCalibrationMethod.ThispaperwaspresentedatICCV2023andfocusesonaddressingthechallengeoflearningfromlong-taileddatawithnoisylabels.

Let’stalkaboutthebackground.Whatarethelong-taileddataandnoisylabels?Pleaseobservetheleftfigureontheslide.Intherealworld,thedatacollectedareimbalanced;someclasseshavealargenumberofsamples,whileothershaveasmallnumber.Withinthedataset,thenumberofsamplesforeachclasstakesontheshapeofalong-tail.

Noisylabelsmeanthelabelmismatcheswithitsdata.Forexample,inthe‘dog’class,acatimageislabeledas‘dog’,thatisa‘noisylabel’.

Inpreviouswork,methodsforlearningwithnoisylabelsincludedthememorizationeffectandthenoisetransitionmatrix.Bothmethodsaresusceptibletotheimbalancedsamplesizesinherentinlong-taileddata.

Similarly,methodsforlearningwithlong-taileddatatypicallyinvolveadaptingre-samplingandre-weightingtechniques.However,thesemethodstendtoaccumulateerrorinformationfromnoisylabels.

Methodsforaddressingbothproblemssimultaneouslycanbecategorizedintotwomainapproaches:

(1)Distinguishingmislabeleddatafromdatabelongingtotailclassesforsubsequentprocedures.

(2)Mitigatingthesideeffectsofmislabeleddataandlong-taileddatainaunifiedmanner,typicallyrelyingonstrongassumptions.

Thefirstmethodnecessitatesrepresentationstodiscernwhetheralabelisnoisy,buttheserepresentationsarederivedfromnoisydata,thuslackingrobustness.

Ontheotherhand,thesecondmethodreliesheavilyonstrongassumptionsandmaynotgeneralizewell.

TheauthorproposedRCALinspiredbythes

您可能关注的文档

文档评论(0)

170333868030ac7 + 关注
实名认证
内容提供者

8065113005000065

1亿VIP精品文档

相关文档