解释性人工智能基准测试 - 关于可用工具包和开放挑战的调查.docx

下载文档

0
0
约4.99万字
约 9页
2024-04-18 发布于境外
举报
版权申诉
保障服务

解释性人工智能基准测试 - 关于可用工具包和开放挑战的调查.docx

1、本文档共9页，可阅读全部内容。
2、原创力文档（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、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、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)。

Proceedings

ProceedingsoftheThirty-SecondInternationalJointConferenceonArtificialIntelligence(IJCAI-23)SurveyTrack

PAGE

PAGE6666

BenchmarkingeXplainableAI-ASurveyonAvailableToolkitsandOpenChallenges

PhuongQuynhLe1,MeikeNauta1,2,VanBachNguyen1,ShreyasiPathak1,2,

Jo¨rgSchlo¨tterer1,3,4andChristinSeifert1,31UniversityofDuisburg-Essen,Germany2UniversityofTwente,theNetherlands3UniversityofMarburg,Germany

4UniversityofMannheim,Germany

{phuongquynh.le,vanbach.nguyen}@uni-due.de,{m.nauta,s.pathak}@utwente.nl

{joerg.schloetterer,christin.seifert}@uni-marburg.de

Abstract

ThegoalofExplainableAI(XAI)istomakethereasoningofamachinelearningmodelaccessibletohumans,suchthatusersofanAIsystemcanevaluateandjudgetheunderlyingmodel.DuetotheblackboxnatureofXAImethodsitis,however,hardtodisentanglethecontributionofamodelandtheexplanationmethodtothefinaloutput.Itmightbeunclearonwhetheranunexpectedoutputiscausedbythemodelortheexplanationmethod.Explanationmodels,therefore,needtobeevalu-atedintechnical(e.g.fidelitytothemodel)anduser-facing(correspondencetodomainknowledge)terms.Arecentsurveyhasidentified29differ-entautomatedapproachestoquantitativelyevalu-ateexplanations.Inthiswork,wetakeanaddi-tionalperspectiveandanalysewhichtoolkitsanddatasetsareavailable.Weinvestigatewhichevalu-ationmetricsareimplementedinthetoolkitsandwhethertheyproducethesameresults.Wefindthatonlyafewaspectsofexplanationqualityarecurrentlycovered,datasetsarerareandevaluationresultsarenotcomparableacrossdifferenttoolkits.OursurveycanserveasaguidefortheXAIcom-munityforidentifyingfuturedirectionsofresearch,andmostnotably,standardisationofevaluation.

Introduction

ExplainableAI(XAI)hasgrownintoitsownresearcharea,mainlyduetotheemergenceofdeeplearning,th