{"id":"https://openalex.org/W7169655500","doi":"https://doi.org/10.1007/978-981-92-3520-9_14","title":"EXACT: Expert Attribution via Causal Tracing for Targeted Knowledge Unlearning in Mixture-of-Experts Language Models","display_name":"EXACT: Expert Attribution via Causal Tracing for Targeted Knowledge Unlearning in Mixture-of-Experts Language Models","publication_year":2026,"publication_date":"2026-07-18","ids":{"openalex":"https://openalex.org/W7169655500","doi":"https://doi.org/10.1007/978-981-92-3520-9_14"},"language":"en","primary_location":{"id":"doi:10.1007/978-981-92-3520-9_14","is_oa":true,"landing_page_url":"https://doi.org/10.1007/978-981-92-3520-9_14","pdf_url":"https://link.springer.com/content/pdf/10.1007/978-981-92-3520-9_14.pdf","source":{"id":"https://openalex.org/S106296714","display_name":"Lecture notes in computer science","issn_l":"0302-9743","issn":["0302-9743","1611-3349"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319900","host_organization_name":"Springer Science+Business Media","host_organization_lineage":["https://openalex.org/P4310319900","https://openalex.org/P4310319965"],"host_organization_lineage_names":["Springer Science+Business Media","Springer Nature"],"type":"book series"},"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Lecture Notes in Computer Science","raw_type":"book-chapter"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://link.springer.com/content/pdf/10.1007/978-981-92-3520-9_14.pdf","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137283058","display_name":"Qicheng Wang","orcid":null},"institutions":[{"id":"https://openalex.org/I149594827","display_name":"Xidian University","ror":"https://ror.org/05s92vm98","country_code":"CN","type":"education","lineage":["https://openalex.org/I149594827"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Qicheng Wang","raw_affiliation_strings":["School of Computer Science and Technology, Xidian University, Xi\u2019an, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computer Science and Technology, Xidian University, Xi\u2019an, China","institution_ids":["https://openalex.org/I149594827"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5141141084","display_name":"Quanhe Wang","orcid":null},"institutions":[{"id":"https://openalex.org/I149594827","display_name":"Xidian University","ror":"https://ror.org/05s92vm98","country_code":"CN","type":"education","lineage":["https://openalex.org/I149594827"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Quanhe Wang","raw_affiliation_strings":["School of Computer Science and Technology, Xidian University, Xi\u2019an, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computer Science and Technology, Xidian University, Xi\u2019an, China","institution_ids":["https://openalex.org/I149594827"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5141141743","display_name":"Lu Wang","orcid":null},"institutions":[{"id":"https://openalex.org/I149594827","display_name":"Xidian University","ror":"https://ror.org/05s92vm98","country_code":"CN","type":"education","lineage":["https://openalex.org/I149594827"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Lu Wang","raw_affiliation_strings":["School of Computer Science and Technology, Xidian University, Xi\u2019an, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computer Science and Technology, Xidian University, Xi\u2019an, China","institution_ids":["https://openalex.org/I149594827"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":["https://openalex.org/A5141141743"],"corresponding_institution_ids":["https://openalex.org/I149594827"],"apc_list":{"value":5000,"currency":"EUR","value_usd":5392},"apc_paid":{"value":5000,"currency":"EUR","value_usd":5392},"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":"170","last_page":"181"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":null,"topics":[],"keywords":[{"id":"https://openalex.org/keywords/forgetting","display_name":"Forgetting","score":0.5160999894142151},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.49889999628067017},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.4602999985218048},{"id":"https://openalex.org/keywords/boosting","display_name":"Boosting (machine learning)","score":0.429500013589859},{"id":"https://openalex.org/keywords/adversarial-system","display_name":"Adversarial system","score":0.3962000012397766},{"id":"https://openalex.org/keywords/knowledge-representation-and-reasoning","display_name":"Knowledge representation and reasoning","score":0.38370001316070557},{"id":"https://openalex.org/keywords/tracing","display_name":"Tracing","score":0.3653999865055084},{"id":"https://openalex.org/keywords/salience","display_name":"Salience (neuroscience)","score":0.3431999981403351}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8421000242233276},{"id":"https://openalex.org/C7149132","wikidata":"https://www.wikidata.org/wiki/Q1377840","display_name":"Forgetting","level":2,"score":0.5160999894142151},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5076000094413757},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.49889999628067017},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.4602999985218048},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.429500013589859},{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.3962000012397766},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.38960000872612},{"id":"https://openalex.org/C161301231","wikidata":"https://www.wikidata.org/wiki/Q3478658","display_name":"Knowledge representation and reasoning","level":2,"score":0.38370001316070557},{"id":"https://openalex.org/C138673069","wikidata":"https://www.wikidata.org/wiki/Q322229","display_name":"Tracing","level":2,"score":0.3653999865055084},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.35100001096725464},{"id":"https://openalex.org/C108154423","wikidata":"https://www.wikidata.org/wiki/Q1469792","display_name":"Salience (neuroscience)","level":2,"score":0.3431999981403351},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.31450000405311584},{"id":"https://openalex.org/C108650721","wikidata":"https://www.wikidata.org/wiki/Q1783253","display_name":"Counterfactual thinking","level":2,"score":0.3124000132083893},{"id":"https://openalex.org/C143299363","wikidata":"https://www.wikidata.org/wiki/Q900584","display_name":"Attribution","level":2,"score":0.30079999566078186},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.29739999771118164},{"id":"https://openalex.org/C2777220311","wikidata":"https://www.wikidata.org/wiki/Q6423340","display_name":"Knowledge acquisition","level":2,"score":0.2816999852657318},{"id":"https://openalex.org/C93226319","wikidata":"https://www.wikidata.org/wiki/Q193137","display_name":"Differential (mechanical device)","level":2,"score":0.27730000019073486},{"id":"https://openalex.org/C2983448237","wikidata":"https://www.wikidata.org/wiki/Q1078276","display_name":"Language understanding","level":2,"score":0.27379998564720154},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.2736999988555908},{"id":"https://openalex.org/C66746571","wikidata":"https://www.wikidata.org/wiki/Q1134833","display_name":"ENCODE","level":3,"score":0.2621000111103058},{"id":"https://openalex.org/C44291984","wikidata":"https://www.wikidata.org/wiki/Q1074173","display_name":"Question answering","level":2,"score":0.25859999656677246},{"id":"https://openalex.org/C4554734","wikidata":"https://www.wikidata.org/wiki/Q593744","display_name":"Knowledge base","level":2,"score":0.25540000200271606}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1007/978-981-92-3520-9_14","is_oa":true,"landing_page_url":"https://doi.org/10.1007/978-981-92-3520-9_14","pdf_url":"https://link.springer.com/content/pdf/10.1007/978-981-92-3520-9_14.pdf","source":{"id":"https://openalex.org/S106296714","display_name":"Lecture notes in computer science","issn_l":"0302-9743","issn":["0302-9743","1611-3349"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319900","host_organization_name":"Springer Science+Business Media","host_organization_lineage":["https://openalex.org/P4310319900","https://openalex.org/P4310319965"],"host_organization_lineage_names":["Springer Science+Business Media","Springer Nature"],"type":"book series"},"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Lecture Notes in Computer Science","raw_type":"book-chapter"}],"best_oa_location":{"id":"doi:10.1007/978-981-92-3520-9_14","is_oa":true,"landing_page_url":"https://doi.org/10.1007/978-981-92-3520-9_14","pdf_url":"https://link.springer.com/content/pdf/10.1007/978-981-92-3520-9_14.pdf","source":{"id":"https://openalex.org/S106296714","display_name":"Lecture notes in computer science","issn_l":"0302-9743","issn":["0302-9743","1611-3349"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319900","host_organization_name":"Springer Science+Business Media","host_organization_lineage":["https://openalex.org/P4310319900","https://openalex.org/P4310319965"],"host_organization_lineage_names":["Springer Science+Business Media","Springer Nature"],"type":"book series"},"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Lecture Notes in Computer Science","raw_type":"book-chapter"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7169655500.pdf","grobid_xml":"https://content.openalex.org/works/W7169655500.grobid-xml"},"referenced_works_count":6,"referenced_works":["https://openalex.org/W1488996941","https://openalex.org/W3154155772","https://openalex.org/W4412945481","https://openalex.org/W4415796002","https://openalex.org/W4415796188","https://openalex.org/W7133236469"],"related_works":[],"abstract_inverted_index":{"Machine":[0],"unlearning":[1,25,64,126,173],"has":[2],"become":[3],"increasingly":[4],"important":[5],"for":[6,62],"removing":[7],"hazardous":[8],"or":[9],"unauthorized":[10],"knowledge":[11,51,187],"from":[12,43,198,202],"deployed":[13],"language":[14,22,67],"models":[15],"without":[16],"full":[17],"retraining.":[18],"In":[19],"Mixture-of-Experts":[20],"(MoE)":[21],"models,":[23],"existing":[24],"methods":[26],"face":[27],"a":[28,59,87,138],"structural":[29],"failure":[30],"mode:":[31],"they":[32],"tend":[33],"to":[34,97,123,137],"reduce":[35],"forget-set":[36],"performance":[37],"by":[38,150],"merely":[39],"redirecting":[40],"queries":[41],"away":[42],"knowledgeable":[44],"experts,":[45],"rather":[46],"than":[47,117,189,201],"erasing":[48],"the":[49,98,111,130,165],"underlying":[50],"stored":[52],"in":[53,65,141,175],"expert":[54,100],"parameters.":[55,121],"We":[56],"present":[57],"EXACT,":[58],"three-stage":[60],"framework":[61],"targeted":[63],"MoE":[66,158],"models.":[68],"EXACT":[69,108,128,169,183],"first":[70],"screens":[71],"candidate":[72],"experts":[73],"using":[74],"differential":[75],"routing":[76,89,203],"statistics,":[77],"then":[78],"verifies":[79],"their":[80],"causal":[81],"responsibility":[82],"through":[83],"expert-level":[84],"interventions":[85],"under":[86,179],"fixed":[88],"path,":[90],"and":[91,160],"finally":[92],"applies":[93],"gradient-masked":[94],"updates":[95],"exclusively":[96],"verified":[99],"feed-forward":[101],"projections":[102],"while":[103,114,144],"keeping":[104],"all":[105],"routers":[106],"frozen.":[107],"consistently":[109],"improves":[110],"forget-utility":[112],"trade-off":[113],"updating":[115],"less":[116],"1%":[118],"of":[119],"model":[120,148],"Compared":[122],"standard":[124],"full-model":[125],"baselines,":[127,190],"approaches":[129,174],"ideal":[131],"random-guess":[132],"reference":[133],"level":[134],"with":[135],"up":[136],"7.5%":[139],"improvement":[140],"forget":[142],"quality,":[143],"simultaneously":[145],"boosting":[146],"retained":[147],"utility":[149],"on":[151],"average":[152],"about":[153],"40%":[154],"relatively":[155],"across":[156],"two":[157,161],"backbones":[159],"hazardous-knowledge":[162],"domains.":[163],"On":[164],"WMDP-Cyber":[166],"benchmark":[167],"specifically,":[168],"outperforms":[170],"prior":[171],"MoE-specific":[172],"most":[176],"configurations.":[177],"Furthermore,":[178],"adversarial":[180],"expert-forcing":[181],"evaluations,":[182],"exhibits":[184],"significantly":[185],"lower":[186],"recovery":[188],"confirming":[191],"that":[192],"EXACT\u2019s":[193],"forgetting":[194],"effect":[195],"stems":[196],"more":[197],"expert-internal":[199],"modification":[200],"manipulation.":[204]},"counts_by_year":[],"updated_date":"2026-08-21T09:56:20.448147","created_date":"2026-07-19T00:00:00"}