CRAFT: Concept Recursive Activation FacTorization for Explainability
Attribution methods, which employ heatmaps to identify the most influential regions of an image that impact model decisions, have gained widespread popularity as a type of explainability method. However, recent research has exposed the limited practical value of these methods, attributed in part to...
Uložené v:
| Vydané v: | Proceedings (IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Online) Ročník 2023; s. 2711 - 2721 |
|---|---|
| Hlavní autori: | , , , , , , , |
| Médium: | Konferenčný príspevok.. Journal Article |
| Jazyk: | English |
| Vydavateľské údaje: |
United States
IEEE
01.06.2023
|
| Predmet: | |
| ISSN: | 1063-6919, 1063-6919 |
| On-line prístup: | Získať plný text |
| Tagy: |
Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
|
| Abstract | Attribution methods, which employ heatmaps to identify the most influential regions of an image that impact model decisions, have gained widespread popularity as a type of explainability method. However, recent research has exposed the limited practical value of these methods, attributed in part to their narrow focus on the most prominent regions of an image - revealing "where" the model looks, but failing to elucidate "what" the model sees in those areas. In this work, we try to fill in this gap with CRAFT - a novel approach to identify both "what" and "where" by generating concept-based explanations. We introduce 3 new ingredients to the automatic concept extraction literature: (i) a recursive strategy to detect and decompose concepts across layers, (ii) a novel method for a more faithful estimation of concept importance using Sobol indices, and (iii) the use of implicit differentiation to unlock Concept Attribution Maps. We conduct both human and computer vision experiments to demonstrate the benefits of the proposed approach. We show that the proposed concept importance estimation technique is more faithful to the model than previous methods. When evaluating the usefulness of the method for human experimenters on a human-centered utility benchmark, we find that our approach significantly improves on two of the three test scenarios. |
|---|---|
| AbstractList | Attribution methods, which employ heatmaps to identify the most influential regions of an image that impact model decisions, have gained widespread popularity as a type of explainability method. However, recent research has exposed the limited practical value of these methods, attributed in part to their narrow focus on the most prominent regions of an image - revealing "where" the model looks, but failing to elucidate "what" the model sees in those areas. In this work, we try to fill in this gap with CRAFT - a novel approach to identify both "what" and "where" by generating concept-based explanations. We introduce 3 new ingredients to the automatic concept extraction literature: (i) a recursive strategy to detect and decompose concepts across layers, (ii) a novel method for a more faithful estimation of concept importance using Sobol indices, and (iii) the use of implicit differentiation to unlock Concept Attribution Maps. We conduct both human and computer vision experiments to demonstrate the benefits of the proposed approach. We show that the proposed concept importance estimation technique is more faithful to the model than previous methods. When evaluating the usefulness of the method for human experimenters on a human-centered utility benchmark, we find that our approach significantly improves on two of the three test scenarios. Attribution methods, which employ heatmaps to identify the most influential regions of an image that impact model decisions, have gained widespread popularity as a type of explainability method. However, recent research has exposed the limited practical value of these methods, attributed in part to their narrow focus on the most prominent regions of an image - revealing "where" the model looks, but failing to elucidate "what" the model sees in those areas. In this work, we try to fill in this gap with CRAFT - a novel approach to identify both "what" and "where" by generating concept-based explanations. We introduce 3 new ingredients to the automatic concept extraction literature: ( ) a recursive strategy to detect and decompose concepts across layers, ( ) a novel method for a more faithful estimation of concept importance using Sobol indices, and ( ) the use of implicit differentiation to unlock Concept Attribution Maps. We conduct both human and computer vision experiments to demonstrate the benefits of the proposed approach. We show that the proposed concept importance estimation technique is more faithful to the model than previous methods. When evaluating the usefulness of the method for human experimenters on a human-centered utility benchmark, we find that our approach significantly improves on two of the three test scenarios. Our code is freely available: github.com/deel-ai/Craft. Attribution methods, which employ heatmaps to identify the most influential regions of an image that impact model decisions, have gained widespread popularity as a type of explainability method. However, recent research has exposed the limited practical value of these methods, attributed in part to their narrow focus on the most prominent regions of an image - revealing "where" the model looks, but failing to elucidate "what" the model sees in those areas. In this work, we try to fill in this gap with CRAFT - a novel approach to identify both "what" and "where" by generating concept-based explanations. We introduce 3 new ingredients to the automatic concept extraction literature: (i) a recursive strategy to detect and decompose concepts across layers, (ii) a novel method for a more faithful estimation of concept importance using Sobol indices, and (iii) the use of implicit differentiation to unlock Concept Attribution Maps. We conduct both human and computer vision experiments to demonstrate the benefits of the proposed approach. We show that the proposed concept importance estimation technique is more faithful to the model than previous methods. When evaluating the usefulness of the method for human experimenters on a human-centered utility benchmark, we find that our approach significantly improves on two of the three test scenarios. Our code is freely available: github.com/deel-ai/Craft.Attribution methods, which employ heatmaps to identify the most influential regions of an image that impact model decisions, have gained widespread popularity as a type of explainability method. However, recent research has exposed the limited practical value of these methods, attributed in part to their narrow focus on the most prominent regions of an image - revealing "where" the model looks, but failing to elucidate "what" the model sees in those areas. In this work, we try to fill in this gap with CRAFT - a novel approach to identify both "what" and "where" by generating concept-based explanations. We introduce 3 new ingredients to the automatic concept extraction literature: (i) a recursive strategy to detect and decompose concepts across layers, (ii) a novel method for a more faithful estimation of concept importance using Sobol indices, and (iii) the use of implicit differentiation to unlock Concept Attribution Maps. We conduct both human and computer vision experiments to demonstrate the benefits of the proposed approach. We show that the proposed concept importance estimation technique is more faithful to the model than previous methods. When evaluating the usefulness of the method for human experimenters on a human-centered utility benchmark, we find that our approach significantly improves on two of the three test scenarios. Our code is freely available: github.com/deel-ai/Craft. |
| Author | Boissin, Thibaut Colin, Julien Fel, Thomas Bethune, Louis Serre, Thomas Cadenc, Remi Picard, Agustin Vigouroux, David |
| Author_xml | – sequence: 1 givenname: Thomas surname: Fel fullname: Fel, Thomas organization: Carney Institute for Brain Science, Brown University,USA – sequence: 2 givenname: Agustin surname: Picard fullname: Picard, Agustin organization: Artificial and Natural Intelligence Toulouse Institute, Université Toulouse,France – sequence: 3 givenname: Louis surname: Bethune fullname: Bethune, Louis organization: Artificial and Natural Intelligence Toulouse Institute, Université Toulouse,France – sequence: 4 givenname: Thibaut surname: Boissin fullname: Boissin, Thibaut organization: Artificial and Natural Intelligence Toulouse Institute, Université Toulouse,France – sequence: 5 givenname: David surname: Vigouroux fullname: Vigouroux, David organization: Artificial and Natural Intelligence Toulouse Institute, Université Toulouse,France – sequence: 6 givenname: Julien surname: Colin fullname: Colin, Julien organization: Carney Institute for Brain Science, Brown University,USA – sequence: 7 givenname: Remi surname: Cadenc fullname: Cadenc, Remi organization: Carney Institute for Brain Science, Brown University,USA – sequence: 8 givenname: Thomas surname: Serre fullname: Serre, Thomas organization: Carney Institute for Brain Science, Brown University,USA |
| BackLink | https://www.ncbi.nlm.nih.gov/pubmed/38463608$$D View this record in MEDLINE/PubMed |
| BookMark | eNpNj09Lw0AUxFep2Fr7DYrk6KX17XvZza63ElsVCkqpXstms4GVNIn5U6yf3kKreJoZ5sfAXLFeURaOsTGHKeeg7-L315XACPUUAWkKgFKesZGOtCIBBBy1OmcDDpImUnPd--f7bNQ0HwBAyLnU6pL1SYWSJKgBe4hXs8X6PojLwrqqDVbOdnXjdy6Y2dbvTOvLIlgYuy5r_31MWVkH868qN74wic99u79mF5nJGzc66ZC9Lebr-GmyfHl8jmfLiSeQ7SQMlUHNrQkzG4VhlKWQoEOKbApSCp4QQaJM4nhEWlo0ShiRkkpDQTxKgYbs9rhb1eVn55p2s_WNdXluCld2zQa1ECgjheqA3pzQLtm6dFPVfmvq_eb3-QEYHwHvnPurOSAIRKIfxPRoow |
| CODEN | IEEPAD |
| ContentType | Conference Proceeding Journal Article |
| DBID | 6IE 6IH CBEJK RIE RIO NPM 7X8 |
| DOI | 10.1109/CVPR52729.2023.00266 |
| DatabaseName | IEEE Electronic Library (IEL) Conference Proceedings IEEE Proceedings Order Plan (POP) 1998-present by volume IEEE Xplore All Conference Proceedings IEEE/IET Electronic Library IEEE Proceedings Order Plans (POP) 1998-present PubMed MEDLINE - Academic |
| DatabaseTitle | PubMed MEDLINE - Academic |
| DatabaseTitleList | PubMed MEDLINE - Academic |
| Database_xml | – sequence: 1 dbid: NPM name: PubMed url: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=PubMed sourceTypes: Index Database – sequence: 2 dbid: RIE name: IEEE/IET Electronic Library (IEL) (UW System Shared) url: https://ieeexplore.ieee.org/ sourceTypes: Publisher – sequence: 3 dbid: 7X8 name: MEDLINE - Academic url: https://search.proquest.com/medline sourceTypes: Aggregation Database |
| DeliveryMethod | fulltext_linktorsrc |
| Discipline | Applied Sciences Computer Science |
| EISBN | 9798350301298 |
| EISSN | 1063-6919 |
| EndPage | 2721 |
| ExternalDocumentID | 38463608 10205223 |
| Genre | orig-research Journal Article |
| GroupedDBID | 6IE 6IH 6IL 6IN AAWTH ABLEC ADZIZ ALMA_UNASSIGNED_HOLDINGS BEFXN BFFAM BGNUA BKEBE BPEOZ CBEJK CHZPO IEGSK IJVOP OCL RIE RIL RIO 23M 29F 29O 6IK ABDPE ACGFS IPLJI M43 NPM RIG RNS 7X8 |
| ID | FETCH-LOGICAL-i306t-448a291ca4fc7447fd0b2e237cd06651b330b8abe17396c2a85a5d38d45317d03 |
| IEDL.DBID | RIE |
| ISICitedReferencesCount | 45 |
| ISICitedReferencesURI | http://www.webofscience.com/api/gateway?GWVersion=2&SrcApp=Summon&SrcAuth=ProQuest&DestLinkType=CitingArticles&DestApp=WOS_CPL&KeyUT=001058542603004&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D |
| ISSN | 1063-6919 |
| IngestDate | Fri Sep 05 06:48:54 EDT 2025 Wed Feb 19 02:12:36 EST 2025 Wed Aug 27 02:56:33 EDT 2025 |
| IsDoiOpenAccess | false |
| IsOpenAccess | true |
| IsPeerReviewed | false |
| IsScholarly | true |
| Language | English |
| LinkModel | DirectLink |
| MergedId | FETCHMERGED-LOGICAL-i306t-448a291ca4fc7447fd0b2e237cd06651b330b8abe17396c2a85a5d38d45317d03 |
| Notes | ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 23 |
| OpenAccessLink | https://pmc.ncbi.nlm.nih.gov/articles/PMC10923575/pdf/nihms-1924466.pdf |
| PMID | 38463608 |
| PQID | 2955267828 |
| PQPubID | 23479 |
| PageCount | 11 |
| ParticipantIDs | proquest_miscellaneous_2955267828 pubmed_primary_38463608 ieee_primary_10205223 |
| PublicationCentury | 2000 |
| PublicationDate | 2023-Jun |
| PublicationDateYYYYMMDD | 2023-06-01 |
| PublicationDate_xml | – month: 06 year: 2023 text: 2023-Jun |
| PublicationDecade | 2020 |
| PublicationPlace | United States |
| PublicationPlace_xml | – name: United States |
| PublicationTitle | Proceedings (IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Online) |
| PublicationTitleAbbrev | CVPR |
| PublicationTitleAlternate | Proc IEEE Comput Soc Conf Comput Vis Pattern Recognit |
| PublicationYear | 2023 |
| Publisher | IEEE |
| Publisher_xml | – name: IEEE |
| SSID | ssj0003211698 ssj0023720 |
| Score | 2.7060275 |
| Snippet | Attribution methods, which employ heatmaps to identify the most influential regions of an image that impact model decisions, have gained widespread popularity... |
| SourceID | proquest pubmed ieee |
| SourceType | Aggregation Database Index Database Publisher |
| StartPage | 2711 |
| SubjectTerms | Benchmark testing Computational modeling Computer vision Decision making Estimation Explainable computer vision Heating systems Pattern recognition |
| Title | CRAFT: Concept Recursive Activation FacTorization for Explainability |
| URI | https://ieeexplore.ieee.org/document/10205223 https://www.ncbi.nlm.nih.gov/pubmed/38463608 https://www.proquest.com/docview/2955267828 |
| Volume | 2023 |
| WOSCitedRecordID | wos001058542603004&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D |
| hasFullText | 1 |
| inHoldings | 1 |
| isFullTextHit | |
| isPrint | |
| link | http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV07T8MwED7RioGpPAqUR2Uk1pTEjl9sKBAxVVVVULfIsR2pS4r6QOLfYztpmTqwZXFkne903z2-O4BHF1RVQlMbGUPKKC2ZjmTl7IopZRSrDK10HJZN8PFYzOdy0pLVAxfGWhuaz-zIf4ZavlnqrU-VOQvHscMLpAMdznlD1tonVIgLZZgULT0uieVT9jmZUuzQ48jvCPepEz8LMSxROYwng1_Je_-80Sn0_xh6aLL3PWdwZOtz6LWQErUGu76A12z6ks-eUdawE9HUp9d9xzp60bvFZihXerZctYRM5FAs8p15gVblO2d_-vCRv82y96hdnBAtXASwiVzIpbBMtEorzdOUVyYuscWEa-MrLUlJSFwKVdqEE8k0VoIqaogwqbNIbmJyCd16WdtrQExKpSzDmrMqtTRVuLR-KD0ufbKX8QH0vVSKr2Y2RrETyAAedgIunML6KoSq7XK7LrCkFDsXicUArhrJ708TEeaXiZsDf72FE_-UTavWHXQ3q629h2P9vVmsV0OnFXMxDFrxC1kstmU |
| linkProvider | IEEE |
| linkToHtml | http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV3JTsMwEB1BQYJTWQqU1UhcUxJvsbmhQlVEqaqqIG6RYztSLy3qgsTfYztpOfXALZdE1sSjebO8eQB3LqkqhGY2MobkEc25jmTh_IorZRQvDCt0HMQm0n5ffH7KQUVWD1wYa20YPrMt_xh6-Waql75U5jwcxw4vkG3YYZTipKRrrUsqxCUzXIqKIJfE8r79MRgy7PBjy6uE--KJ34YYZFQ2I8oQWTr1f57pABp_HD00WEefQ9iykyOoV6ASVS47P4an9vCxM3pA7ZKfiIa-wO5n1tGjXkmboY7So-msomQih2ORn80LxCo_O_vTgPfO86jdjSrphGjscoBF5JIuhWWiFS10SmlamDjHFpNUG99rSXJC4lyo3CYpkVxjJZhihghDnU-mJiYnUJtMJ_YMEJdSKcuxTnlBLaMK59avpce5L_fytAkNb5Xsq9yOka0M0oTblYEzd2V9H0JN7HQ5z7BkDLsgiUUTTkvLr98mImwwE-cbvnoDe93RWy_rvfRfL2Df_9ZycOsSaovZ0l7Brv5ejOez63A3fgG0-LjE |
| openUrl | ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=proceeding&rft.title=Proceedings+%28IEEE+Computer+Society+Conference+on+Computer+Vision+and+Pattern+Recognition.+Online%29&rft.atitle=CRAFT%3A+Concept+Recursive+Activation+FacTorization+for+Explainability&rft.au=Fel%2C+Thomas&rft.au=Picard%2C+Agustin&rft.au=Bethune%2C+Louis&rft.au=Boissin%2C+Thibaut&rft.date=2023-06-01&rft.pub=IEEE&rft.eissn=1063-6919&rft.spage=2711&rft.epage=2721&rft_id=info:doi/10.1109%2FCVPR52729.2023.00266&rft.externalDocID=10205223 |
| thumbnail_l | http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=1063-6919&client=summon |
| thumbnail_m | http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=1063-6919&client=summon |
| thumbnail_s | http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=1063-6919&client=summon |