PredDiff: Explanations and interactions from conditional expectations
PredDiff is a model-agnostic, local attribution method that is firmly rooted in probability theory. Its simple intuition is to measure prediction changes while marginalizing features. In this work, we clarify properties of PredDiff and its close connection to Shapley values. We stress important diff...
Gespeichert in:
| Veröffentlicht in: | Artificial intelligence 2022-11, Vol.312, p.103774, Article 103774 |
|---|---|
| Hauptverfasser: | , , |
| Format: | Artikel |
| Sprache: | eng |
| Schlagworte: | |
| Online-Zugang: | Volltext |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| container_end_page | |
|---|---|
| container_issue | |
| container_start_page | 103774 |
| container_title | Artificial intelligence |
| container_volume | 312 |
| creator | Blücher, Stefan Vielhaben, Johanna Strodthoff, Nils |
| description | PredDiff is a model-agnostic, local attribution method that is firmly rooted in probability theory. Its simple intuition is to measure prediction changes while marginalizing features. In this work, we clarify properties of PredDiff and its close connection to Shapley values. We stress important differences between classification and regression, which require a specific treatment within both formalisms. We extend PredDiff by introducing a new, well-founded measure for interaction effects between arbitrary feature subsets. The study of interaction effects represents an inevitable step towards a comprehensive understanding of black-box models and is particularly important for science applications. Equipped with our novel interaction measure, PredDiff is a promising model-agnostic approach for obtaining reliable, numerically inexpensive and theoretically sound attributions. |
| doi_str_mv | 10.1016/j.artint.2022.103774 |
| format | Article |
| fullrecord | <record><control><sourceid>proquest_cross</sourceid><recordid>TN_cdi_proquest_journals_2760225527</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><els_id>S000437022200114X</els_id><sourcerecordid>2760225527</sourcerecordid><originalsourceid>FETCH-LOGICAL-c380t-e5850d37f3e01faa22a3d7c66b611661e696221e5fe1c9e3f85c0102607f16773</originalsourceid><addsrcrecordid>eNp9UE1LAzEUDKJgrf4DDwuet-Yl3WTXgyC1fkBBD3oOMXmBLO3ummyl_vtmiWdPjxlmhjdDyDXQBVAQt-1Ch9F344JRxhLFpVyekBnUkpWyYXBKZpTSZcklZefkIsY2Qd40MCPr94D20Tt3V6wPw1Z3evR9Fwvd2SIlYtAmEy70u8L0nfUT1tsCDwOaMcsvyZnT24hXf3dOPp_WH6uXcvP2_Lp62JSG13Qssaorarl0HCk4rRnT3EojxJcAEAJQNIIxwMohmAa5qytDgTJBpQMhJZ-Tm5w7hP57j3FUbb8P6ZuomBSpfFWxSbXMKhP6GAM6NQS_0-FXAVXTYKpVeTA1DabyYMl2n22YGvx4DCoaj51B60Nqqmzv_w84Al0MdVs</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2760225527</pqid></control><display><type>article</type><title>PredDiff: Explanations and interactions from conditional expectations</title><source>Elsevier ScienceDirect Journals</source><creator>Blücher, Stefan ; Vielhaben, Johanna ; Strodthoff, Nils</creator><creatorcontrib>Blücher, Stefan ; Vielhaben, Johanna ; Strodthoff, Nils</creatorcontrib><description>PredDiff is a model-agnostic, local attribution method that is firmly rooted in probability theory. Its simple intuition is to measure prediction changes while marginalizing features. In this work, we clarify properties of PredDiff and its close connection to Shapley values. We stress important differences between classification and regression, which require a specific treatment within both formalisms. We extend PredDiff by introducing a new, well-founded measure for interaction effects between arbitrary feature subsets. The study of interaction effects represents an inevitable step towards a comprehensive understanding of black-box models and is particularly important for science applications. Equipped with our novel interaction measure, PredDiff is a promising model-agnostic approach for obtaining reliable, numerically inexpensive and theoretically sound attributions.</description><identifier>ISSN: 0004-3702</identifier><identifier>EISSN: 1872-7921</identifier><identifier>DOI: 10.1016/j.artint.2022.103774</identifier><language>eng</language><publisher>Amsterdam: Elsevier B.V</publisher><subject>Explainable AI ; Feature attribution ; Interactions ; Interpretability ; Mathematical models ; Probability theory ; Shapley values ; Statistical analysis</subject><ispartof>Artificial intelligence, 2022-11, Vol.312, p.103774, Article 103774</ispartof><rights>2022 The Authors</rights><rights>Copyright Elsevier Science Ltd. Nov 2022</rights><lds50>peer_reviewed</lds50><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed><citedby>FETCH-LOGICAL-c380t-e5850d37f3e01faa22a3d7c66b611661e696221e5fe1c9e3f85c0102607f16773</citedby><cites>FETCH-LOGICAL-c380t-e5850d37f3e01faa22a3d7c66b611661e696221e5fe1c9e3f85c0102607f16773</cites><orcidid>0000-0002-6330-7996 ; 0000-0003-4447-0162</orcidid></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktohtml>$$Uhttps://www.sciencedirect.com/science/article/pii/S000437022200114X$$EHTML$$P50$$Gelsevier$$Hfree_for_read</linktohtml><link.rule.ids>314,776,780,3537,27903,27904,65309</link.rule.ids></links><search><creatorcontrib>Blücher, Stefan</creatorcontrib><creatorcontrib>Vielhaben, Johanna</creatorcontrib><creatorcontrib>Strodthoff, Nils</creatorcontrib><title>PredDiff: Explanations and interactions from conditional expectations</title><title>Artificial intelligence</title><description>PredDiff is a model-agnostic, local attribution method that is firmly rooted in probability theory. Its simple intuition is to measure prediction changes while marginalizing features. In this work, we clarify properties of PredDiff and its close connection to Shapley values. We stress important differences between classification and regression, which require a specific treatment within both formalisms. We extend PredDiff by introducing a new, well-founded measure for interaction effects between arbitrary feature subsets. The study of interaction effects represents an inevitable step towards a comprehensive understanding of black-box models and is particularly important for science applications. Equipped with our novel interaction measure, PredDiff is a promising model-agnostic approach for obtaining reliable, numerically inexpensive and theoretically sound attributions.</description><subject>Explainable AI</subject><subject>Feature attribution</subject><subject>Interactions</subject><subject>Interpretability</subject><subject>Mathematical models</subject><subject>Probability theory</subject><subject>Shapley values</subject><subject>Statistical analysis</subject><issn>0004-3702</issn><issn>1872-7921</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2022</creationdate><recordtype>article</recordtype><recordid>eNp9UE1LAzEUDKJgrf4DDwuet-Yl3WTXgyC1fkBBD3oOMXmBLO3ummyl_vtmiWdPjxlmhjdDyDXQBVAQt-1Ch9F344JRxhLFpVyekBnUkpWyYXBKZpTSZcklZefkIsY2Qd40MCPr94D20Tt3V6wPw1Z3evR9Fwvd2SIlYtAmEy70u8L0nfUT1tsCDwOaMcsvyZnT24hXf3dOPp_WH6uXcvP2_Lp62JSG13Qssaorarl0HCk4rRnT3EojxJcAEAJQNIIxwMohmAa5qytDgTJBpQMhJZ-Tm5w7hP57j3FUbb8P6ZuomBSpfFWxSbXMKhP6GAM6NQS_0-FXAVXTYKpVeTA1DabyYMl2n22YGvx4DCoaj51B60Nqqmzv_w84Al0MdVs</recordid><startdate>202211</startdate><enddate>202211</enddate><creator>Blücher, Stefan</creator><creator>Vielhaben, Johanna</creator><creator>Strodthoff, Nils</creator><general>Elsevier B.V</general><general>Elsevier Science Ltd</general><scope>6I.</scope><scope>AAFTH</scope><scope>AAYXX</scope><scope>CITATION</scope><scope>7SC</scope><scope>8FD</scope><scope>JQ2</scope><scope>L7M</scope><scope>L~C</scope><scope>L~D</scope><orcidid>https://orcid.org/0000-0002-6330-7996</orcidid><orcidid>https://orcid.org/0000-0003-4447-0162</orcidid></search><sort><creationdate>202211</creationdate><title>PredDiff: Explanations and interactions from conditional expectations</title><author>Blücher, Stefan ; Vielhaben, Johanna ; Strodthoff, Nils</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-c380t-e5850d37f3e01faa22a3d7c66b611661e696221e5fe1c9e3f85c0102607f16773</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2022</creationdate><topic>Explainable AI</topic><topic>Feature attribution</topic><topic>Interactions</topic><topic>Interpretability</topic><topic>Mathematical models</topic><topic>Probability theory</topic><topic>Shapley values</topic><topic>Statistical analysis</topic><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Blücher, Stefan</creatorcontrib><creatorcontrib>Vielhaben, Johanna</creatorcontrib><creatorcontrib>Strodthoff, Nils</creatorcontrib><collection>ScienceDirect Open Access Titles</collection><collection>Elsevier:ScienceDirect:Open Access</collection><collection>CrossRef</collection><collection>Computer and Information Systems Abstracts</collection><collection>Technology Research Database</collection><collection>ProQuest Computer Science Collection</collection><collection>Advanced Technologies Database with Aerospace</collection><collection>Computer and Information Systems Abstracts Academic</collection><collection>Computer and Information Systems Abstracts Professional</collection><jtitle>Artificial intelligence</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Blücher, Stefan</au><au>Vielhaben, Johanna</au><au>Strodthoff, Nils</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>PredDiff: Explanations and interactions from conditional expectations</atitle><jtitle>Artificial intelligence</jtitle><date>2022-11</date><risdate>2022</risdate><volume>312</volume><spage>103774</spage><pages>103774-</pages><artnum>103774</artnum><issn>0004-3702</issn><eissn>1872-7921</eissn><abstract>PredDiff is a model-agnostic, local attribution method that is firmly rooted in probability theory. Its simple intuition is to measure prediction changes while marginalizing features. In this work, we clarify properties of PredDiff and its close connection to Shapley values. We stress important differences between classification and regression, which require a specific treatment within both formalisms. We extend PredDiff by introducing a new, well-founded measure for interaction effects between arbitrary feature subsets. The study of interaction effects represents an inevitable step towards a comprehensive understanding of black-box models and is particularly important for science applications. Equipped with our novel interaction measure, PredDiff is a promising model-agnostic approach for obtaining reliable, numerically inexpensive and theoretically sound attributions.</abstract><cop>Amsterdam</cop><pub>Elsevier B.V</pub><doi>10.1016/j.artint.2022.103774</doi><orcidid>https://orcid.org/0000-0002-6330-7996</orcidid><orcidid>https://orcid.org/0000-0003-4447-0162</orcidid><oa>free_for_read</oa></addata></record> |
| fulltext | fulltext |
| identifier | ISSN: 0004-3702 |
| ispartof | Artificial intelligence, 2022-11, Vol.312, p.103774, Article 103774 |
| issn | 0004-3702 1872-7921 |
| language | eng |
| recordid | cdi_proquest_journals_2760225527 |
| source | Elsevier ScienceDirect Journals |
| subjects | Explainable AI Feature attribution Interactions Interpretability Mathematical models Probability theory Shapley values Statistical analysis |
| title | PredDiff: Explanations and interactions from conditional expectations |
| url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-28T04%3A52%3A29IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest_cross&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=PredDiff:%20Explanations%20and%20interactions%20from%20conditional%20expectations&rft.jtitle=Artificial%20intelligence&rft.au=Bl%C3%BCcher,%20Stefan&rft.date=2022-11&rft.volume=312&rft.spage=103774&rft.pages=103774-&rft.artnum=103774&rft.issn=0004-3702&rft.eissn=1872-7921&rft_id=info:doi/10.1016/j.artint.2022.103774&rft_dat=%3Cproquest_cross%3E2760225527%3C/proquest_cross%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2760225527&rft_id=info:pmid/&rft_els_id=S000437022200114X&rfr_iscdi=true |