{"id":"https://openalex.org/W7140103919","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.271","title":"PATCH: Mitigating PII Leakage in Language Models with Privacy-Aware Targeted Circuit PatcHing","display_name":"PATCH: Mitigating PII Leakage in Language Models with Privacy-Aware Targeted Circuit PatcHing","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7140103919","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.271"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-eacl.271","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.271","pdf_url":"https://aclanthology.org/2026.findings-eacl.271.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EACL 2026","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-eacl.271.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Anthony Hughes","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Anthony Hughes","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Vasisht Duddu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vasisht Duddu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"N. Asokan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"N. Asokan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Nikolaos Aletras","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Nikolaos Aletras","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":null,"display_name":"Ning Ma","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ning Ma","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.3007721,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"5139","last_page":"5153"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11424","display_name":"Security and Verification in Computing","score":0.2590000033378601,"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/T11424","display_name":"Security and Verification in Computing","score":0.2590000033378601,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.16179999709129333,"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/T12122","display_name":"Physical Unclonable Functions (PUFs) and Hardware Security","score":0.09889999777078629,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/leakage","display_name":"Leakage (economics)","score":0.6169999837875366},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.3346000015735626},{"id":"https://openalex.org/keywords/modeling-language","display_name":"Modeling language","score":0.2678999900817871},{"id":"https://openalex.org/keywords/mathematical-model","display_name":"Mathematical model","score":0.2517000138759613},{"id":"https://openalex.org/keywords/data-modeling","display_name":"Data modeling","score":0.24210000038146973}],"concepts":[{"id":"https://openalex.org/C2777042071","wikidata":"https://www.wikidata.org/wiki/Q6509304","display_name":"Leakage (economics)","level":2,"score":0.6169999837875366},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5203999876976013},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.3346000015735626},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.321399986743927},{"id":"https://openalex.org/C179603123","wikidata":"https://www.wikidata.org/wiki/Q1941921","display_name":"Modeling language","level":3,"score":0.2678999900817871},{"id":"https://openalex.org/C24326235","wikidata":"https://www.wikidata.org/wiki/Q126095","display_name":"Electronic engineering","level":1,"score":0.2556999921798706},{"id":"https://openalex.org/C76969082","wikidata":"https://www.wikidata.org/wiki/Q486902","display_name":"Mathematical model","level":2,"score":0.2517000138759613},{"id":"https://openalex.org/C200601418","wikidata":"https://www.wikidata.org/wiki/Q2193887","display_name":"Reliability engineering","level":1,"score":0.24619999527931213},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.24210000038146973},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.2321999967098236}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-eacl.271","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.271","pdf_url":"https://aclanthology.org/2026.findings-eacl.271.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EACL 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-eacl.271","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.271","pdf_url":"https://aclanthology.org/2026.findings-eacl.271.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G2831959038","display_name":null,"funder_award_id":"EP/S023062","funder_id":"https://openalex.org/F4320314731","funder_display_name":"UK Research and Innovation"},{"id":"https://openalex.org/G5306878388","display_name":null,"funder_award_id":"EP/S023062/1","funder_id":"https://openalex.org/F4320314731","funder_display_name":"UK Research and Innovation"},{"id":"https://openalex.org/G8644793782","display_name":null,"funder_award_id":"EP/S023062/1","funder_id":"https://openalex.org/F4320320342","funder_display_name":"University of Sheffield"}],"funders":[{"id":"https://openalex.org/F4320314607","display_name":"Government of Ontario","ror":"https://ror.org/015pzp858"},{"id":"https://openalex.org/F4320314731","display_name":"UK Research and Innovation","ror":"https://ror.org/001aqnf71"},{"id":"https://openalex.org/F4320320342","display_name":"University of Sheffield","ror":"https://ror.org/05krs5044"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7140103919.pdf","grobid_xml":"https://content.openalex.org/works/W7140103919.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Language":[0],"models":[1],"(LMs)":[2],"may":[3],"memorize":[4],"personally":[5],"identifiable":[6],"information":[7],"(PII)":[8],"from":[9,102],"training":[10],"data,":[11],"enabling":[12],"adversaries":[13],"to":[14,42,86,106,114,123],"extract":[15],"it":[16],"during":[17],"inference.Existing":[18],"defense":[19,141],"mechanisms":[20],"such":[21],"as":[22,124,126],"differential":[23],"privacy":[24],"(DP)":[25],"reduce":[26,87,115],"this":[27,65],"leakage,":[28],"but":[29],"incur":[30],"large":[31],"drops":[32],"in":[33,50,59],"utility.Based":[34],"on":[35],"a":[36,74],"comprehensive":[37],"study":[38],"using":[39],"circuit":[40],"discovery":[41],"identify":[43],"the":[44,137],"computational":[45],"circuits":[46,58,85,133],"responsible":[47,63],"PII":[48,56,84,100,131],"leakage":[49,57,101,119,132],"LMs,":[51],"we":[52,67],"hypothesize":[53],"that":[54,77,130],"specific":[55],"LMs":[60,103],"should":[61],"be":[62,110],"for":[64],"behavior.Therefore,":[66],"propose":[68],"PATCH":[69,108,144],"(Privacy-Aware":[70],"Targeted":[71],"Circuit":[72],"PatcHing),":[73],"novel":[75],"approach":[76],"first":[78],"identifies":[79],"and":[80],"subsequently":[81],"directly":[82],"edits":[83],"leakage.PATCH":[88],"achieves":[89],"better":[90],"privacy-utility":[91],"trade-off":[92],"than":[93],"existing":[94,140],"defenses,":[95],"e.g.,":[96],"reducing":[97],"recall":[98,116],"of":[99,117,120,139],"by":[104],"up":[105],"65%.Finally,":[107],"can":[109,145],"combined":[111],"with":[112],"DP":[113],"residual":[118],"an":[121],"LM":[122],"low":[125],"0.01%.Our":[127],"analysis":[128],"shows":[129],"persist":[134],"even":[135],"after":[136],"application":[138],"mechanisms.In":[142],"contrast,":[143],"effectively":[146],"mitigate":[147],"their":[148],"impact.":[149],"1":[150]},"counts_by_year":[],"updated_date":"2026-03-25T13:04:00.132906","created_date":"2026-03-01T00:00:00"}