{"id":"https://openalex.org/W7220680748","doi":"https://doi.org/10.48550/arxiv.2610.06285","title":"When Are Concept Bottleneck Model Explanations Faithful and Compact?","display_name":"When Are Concept Bottleneck Model Explanations Faithful and Compact?","publication_year":2026,"publication_date":"2026-10-05","ids":{"openalex":"https://openalex.org/W7220680748","doi":"https://doi.org/10.48550/arxiv.2610.06285"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.06285","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.06285","pdf_url":"https://arxiv.org/pdf/2610.06285","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2610.06285","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5001174680","display_name":"Stefano Teso","orcid":"https://orcid.org/0000-0002-2340-9461","observed_orcids":["https://orcid.org/0000-0002-2340-9461"]},"institutions":[{"id":"https://openalex.org/I193223587","display_name":"University of Trento","ror":"https://ror.org/05trd4x28","country_code":"IT","type":"education","lineage":["https://openalex.org/I193223587"]}],"countries":["IT"],"is_corresponding":false,"raw_author_name":"Teso, Stefano","raw_affiliation_strings":["CIMeC & DISI University of Trento"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"CIMeC & DISI University of Trento","institution_ids":["https://openalex.org/I193223587"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5086308196","display_name":"Emanuele Marconato","orcid":"https://orcid.org/0000-0002-7407-5465","observed_orcids":["https://orcid.org/0000-0002-7407-5465"]},"institutions":[{"id":"https://openalex.org/I124055696","display_name":"University of Copenhagen","ror":"https://ror.org/035b05819","country_code":"DK","type":"education","lineage":["https://openalex.org/I124055696"]}],"countries":["DK"],"is_corresponding":false,"raw_author_name":"Marconato, Emanuele","raw_affiliation_strings":["Department of Mathematical Sciences University of Copenhagen"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Mathematical Sciences University of Copenhagen","institution_ids":["https://openalex.org/I124055696"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5080390202","display_name":"Steve Azzolin","orcid":"https://orcid.org/0009-0005-3418-0585","observed_orcids":["https://orcid.org/0009-0005-3418-0585"]},"institutions":[{"id":"https://openalex.org/I193223587","display_name":"University of Trento","ror":"https://ror.org/05trd4x28","country_code":"IT","type":"education","lineage":["https://openalex.org/I193223587"]}],"countries":["IT"],"is_corresponding":false,"raw_author_name":"Azzolin, Steve","raw_affiliation_strings":["DISI University of Trento"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"DISI University of Trento","institution_ids":["https://openalex.org/I193223587"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5069110696","display_name":"Antonio Vergari","orcid":"https://orcid.org/0000-0003-0036-5678","observed_orcids":["https://orcid.org/0000-0003-0036-5678"]},"institutions":[{"id":"https://openalex.org/I98677209","display_name":"University of Edinburgh","ror":"https://ror.org/01nrxwf90","country_code":"GB","type":"education","lineage":["https://openalex.org/I98677209"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Vergari, Antonio","raw_affiliation_strings":["School of Informatics University of Edinburgh"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Informatics University of Edinburgh","institution_ids":["https://openalex.org/I98677209"]}]}],"institutions":[],"countries_distinct_count":3,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":1.0,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":1.0,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10743","display_name":"Software Testing and Debugging Techniques","score":0.0,"subfield":{"id":"https://openalex.org/subfields/1712","display_name":"Software"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.0,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/concept-bottleneck-models","display_name":"concept bottleneck models","score":0.9940000176429749},{"id":"https://openalex.org/keywords/interpretability","display_name":"interpretability","score":0.7289999723434448},{"id":"https://openalex.org/keywords/vision-language-models","display_name":"vision-language models","score":0.6299999952316284}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.9333000183105469},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.7296000123023987},{"id":"https://openalex.org/C5274069","wikidata":"https://www.wikidata.org/wiki/Q2285707","display_name":"Categorical variable","level":2,"score":0.7149999737739563},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.6935999989509583},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5960000157356262},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5789999961853027},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.5091999769210815},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.462799996137619},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.4088999927043915},{"id":"https://openalex.org/C48372109","wikidata":"https://www.wikidata.org/wiki/Q3913","display_name":"Binary number","level":2,"score":0.35740000009536743},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.3555000126361847},{"id":"https://openalex.org/C2776608160","wikidata":"https://www.wikidata.org/wiki/Q4785462","display_name":"Natural (archaeology)","level":2,"score":0.3443000018596649},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.30090001225471497},{"id":"https://openalex.org/C2985583900","wikidata":"https://www.wikidata.org/wiki/Q722617","display_name":"Formal description","level":2,"score":0.296099990606308},{"id":"https://openalex.org/C37616216","wikidata":"https://www.wikidata.org/wiki/Q3218363","display_name":"Lasso (programming language)","level":2,"score":0.2824000120162964},{"id":"https://openalex.org/C60008888","wikidata":"https://www.wikidata.org/wiki/Q6031013","display_name":"Information bottleneck method","level":3,"score":0.2786000072956085}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.06285","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.06285","pdf_url":"https://arxiv.org/pdf/2610.06285","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2610.06285","doi":"https://doi.org/10.48550/arxiv.2610.06285","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.06285","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-nc-sa","license_id":"https://openalex.org/licenses/cc-by-nc-sa","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2610.06285","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.06285","pdf_url":"https://arxiv.org/pdf/2610.06285","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G2070155107","display_name":"It takes two to tango: a synergistic approach to human-machine decision making","funder_award_id":"101120763","funder_id":"https://openalex.org/F4320320300","funder_display_name":"European Commission"},{"id":"https://openalex.org/G2397920","display_name":null,"funder_award_id":"EP/Y023838/1","funder_id":"https://openalex.org/F4320314731","funder_display_name":"UK Research and Innovation"},{"id":"https://openalex.org/G5495295850","display_name":"UNREAL: A Unified Reasoning Layer for Trustworthy ML","funder_award_id":"EP/Y023838/1","funder_id":"https://openalex.org/F4320334627","funder_display_name":"Engineering and Physical Sciences Research Council"},{"id":"https://openalex.org/G5740728094","display_name":null,"funder_award_id":"101120763","funder_id":"https://openalex.org/F8712659396","funder_display_name":"European Health and Digital Executive Agency"},{"id":"https://openalex.org/G6489778452","display_name":null,"funder_award_id":"101120763","funder_id":"https://openalex.org/F4320334322","funder_display_name":"HORIZON EUROPE Framework Programme"}],"funders":[{"id":"https://openalex.org/F4320314731","display_name":"UK Research and Innovation","ror":"https://ror.org/001aqnf71"},{"id":"https://openalex.org/F4320320300","display_name":"European Commission","ror":"https://ror.org/00k4n6c32"},{"id":"https://openalex.org/F4320334322","display_name":"HORIZON EUROPE Framework Programme","ror":null},{"id":"https://openalex.org/F4320334627","display_name":"Engineering and Physical Sciences Research Council","ror":"https://ror.org/0439y7842"},{"id":"https://openalex.org/F8712659396","display_name":"European Health and Digital Executive Agency","ror":"https://ror.org/03xxypa75"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7220680748.pdf","grobid_xml":"https://content.openalex.org/works/W7220680748.grobid-xml"},"referenced_works_count":7,"referenced_works":["https://openalex.org/W2785327160","https://openalex.org/W3173476189","https://openalex.org/W4403995614","https://openalex.org/W6887926099","https://openalex.org/W7161699435","https://openalex.org/W7167078970","https://openalex.org/W7204450287"],"related_works":["https://openalex.org/W7134828437","https://openalex.org/W7124951986","https://openalex.org/W4319654054","https://openalex.org/W4379474664","https://openalex.org/W7163130377","https://openalex.org/W7204978180","https://openalex.org/W4409363398","https://openalex.org/W7133243647","https://openalex.org/W7131859840","https://openalex.org/W4415087864"],"abstract_inverted_index":{"Concept":[0],"bottleneck":[1],"models":[2],"(CBMs)":[3],"are":[4,25,160],"neural":[5],"classifiers":[6],"that":[7],"allow":[8],"to":[9,76,127],"explain":[10],"their":[11,23],"decisions":[12],"via":[13,112],"high-level":[14],"concepts,":[15],"potentially":[16],"enabling":[17],"understanding,":[18],"steering":[19],"and":[20,79,106,129,139,150,154],"debugging.":[21],"However,":[22],"explanations":[24,59],"often":[26],"derived":[27],"heuristically.":[28],"Building":[29],"on":[30],"formal":[31,81,125,152],"explainability,":[32],"we":[33,91],"argue":[34],"they":[35,131],"should":[36],"also":[37,121],"be":[38],"faithful,":[39],"i.e.,":[40],"not":[41],"misreport":[42],"which":[43],"concepts":[44,63,95],"actually":[45],"matter.":[46],"We":[47,120],"show":[48,130],"that,":[49],"for":[50,157],"widespread":[51],"CBM":[52],"architectures,":[53],"including":[54],"recent":[55],"VLM-based":[56],"variants,":[57],"faithful":[58,89],"must":[60],"include":[61],"all":[62],"in":[64,135],"the":[65,85],"bottleneck,":[66],"compromising":[67],"interpretability":[68,148],"when":[69],"this":[70],"is":[71],"large.":[72],"This":[73],"result":[74],"applies":[75],"both":[77],"heuristic":[78],"faithful-by-construction":[80],"explanations.":[82],"To":[83],"encourage":[84],"existence":[86],"of":[87,137],"compact":[88],"explanations,":[90],"suggest":[92],"i)":[93],"modeling":[94],"probabilistically":[96],"as":[97,161,163],"binary":[98],"or":[99],"categorical":[100],"random":[101],"variables":[102],"(rather":[103,115],"than":[104,116],"logits),":[105],"ii)":[107],"employing":[108],"per-concept":[109],"training-time":[110],"sparsification":[111],"group":[113],"lasso":[114],"regular":[117],"elastic":[118],"net).":[119],"extend":[122],"algorithms":[123],"from":[124],"explainability":[126],"CBMs,":[128],"outperform":[132],"natural":[133],"heuristics":[134],"terms":[136],"guarantees":[138],"explanation":[140],"size.":[141],"Overall,":[142],"our":[143],"work":[144],"warns":[145],"against":[146],"naive":[147],"claims":[149],"provides":[151],"conditions":[153],"practical":[155],"strategies":[156],"ensuring":[158],"CBMs":[159],"interpretable":[162],"advertised.":[164]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-07T00:00:00"}