{"id":"https://openalex.org/W4409157766","doi":"https://doi.org/10.1145/3690624.3709312","title":"LLM-Eraser: Optimizing Large Language Model Unlearning through Selective Pruning","display_name":"LLM-Eraser: Optimizing Large Language Model Unlearning through Selective Pruning","publication_year":2025,"publication_date":"2025-04-04","ids":{"openalex":"https://openalex.org/W4409157766","doi":"https://doi.org/10.1145/3690624.3709312"},"language":"en","primary_location":{"id":"doi:10.1145/3690624.3709312","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3690624.3709312","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5085419775","display_name":"Shengming Zhang","orcid":"https://orcid.org/0009-0001-4549-3262"},"institutions":[{"id":"https://openalex.org/I200296433","display_name":"Chinese Academy of Medical Sciences & Peking Union Medical College","ror":"https://ror.org/02drdmm93","country_code":"CN","type":"education","lineage":["https://openalex.org/I200296433"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Shengming Zhang","raw_affiliation_strings":["Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China"],"affiliations":[{"raw_affiliation_string":"Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China","institution_ids":["https://openalex.org/I200296433"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100350641","display_name":"Le Zhang","orcid":"https://orcid.org/0000-0003-0894-9651"},"institutions":[{"id":"https://openalex.org/I98301712","display_name":"Baidu (China)","ror":"https://ror.org/03vs3wt56","country_code":"CN","type":"company","lineage":["https://openalex.org/I98301712"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Le Zhang","raw_affiliation_strings":["Baidu Research, Beijing, China"],"affiliations":[{"raw_affiliation_string":"Baidu Research, Beijing, China","institution_ids":["https://openalex.org/I98301712"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101481194","display_name":"Jingbo Zhou","orcid":"https://orcid.org/0000-0003-2677-7021"},"institutions":[{"id":"https://openalex.org/I98301712","display_name":"Baidu (China)","ror":"https://ror.org/03vs3wt56","country_code":"CN","type":"company","lineage":["https://openalex.org/I98301712"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jingbo Zhou","raw_affiliation_strings":["Baidu Research, Beijing, China"],"affiliations":[{"raw_affiliation_string":"Baidu Research, Beijing, China","institution_ids":["https://openalex.org/I98301712"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5040762883","display_name":"Zhi Zheng","orcid":"https://orcid.org/0000-0001-7758-8904"},"institutions":[{"id":"https://openalex.org/I126520041","display_name":"University of Science and Technology of China","ror":"https://ror.org/04c4dkn09","country_code":"CN","type":"education","lineage":["https://openalex.org/I126520041","https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhi Zheng","raw_affiliation_strings":["University of Science and Technology of China, Hefei, Anhui, China"],"affiliations":[{"raw_affiliation_string":"University of Science and Technology of China, Hefei, Anhui, China","institution_ids":["https://openalex.org/I126520041"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5101862104","display_name":"Hui Xiong","orcid":"https://orcid.org/0000-0001-6016-6465"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hui Xiong","raw_affiliation_strings":["Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China"],"affiliations":[{"raw_affiliation_string":"Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":5,"corresponding_author_ids":["https://openalex.org/A5085419775"],"corresponding_institution_ids":["https://openalex.org/I200296433"],"apc_list":null,"apc_paid":null,"fwci":5.6828,"has_fulltext":false,"cited_by_count":2,"citation_normalized_percentile":{"value":0.95049409,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":91,"max":99},"biblio":{"volume":null,"issue":null,"first_page":"1960","last_page":"1971"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9994000196456909,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.9994000196456909,"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/T10028","display_name":"Topic Modeling","score":0.9987999796867371,"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/T12072","display_name":"Machine Learning and Algorithms","score":0.9914000034332275,"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/pruning","display_name":"Pruning","score":0.6648519039154053},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.6542066335678101},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.4866939187049866},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.47828564047813416},{"id":"https://openalex.org/keywords/natural-language-processing","display_name":"Natural language processing","score":0.4433518648147583},{"id":"https://openalex.org/keywords/speech-recognition","display_name":"Speech recognition","score":0.3287421464920044}],"concepts":[{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.6648519039154053},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6542066335678101},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4866939187049866},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.47828564047813416},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.4433518648147583},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.3287421464920044},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.0},{"id":"https://openalex.org/C6557445","wikidata":"https://www.wikidata.org/wiki/Q173113","display_name":"Agronomy","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3690624.3709312","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3690624.3709312","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[{"display_name":"Quality Education","score":0.550000011920929,"id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":31,"referenced_works":["https://openalex.org/W1488996941","https://openalex.org/W2623399293","https://openalex.org/W2890894339","https://openalex.org/W2946609015","https://openalex.org/W2990138404","https://openalex.org/W2998617917","https://openalex.org/W3013843954","https://openalex.org/W3154155772","https://openalex.org/W3209100754","https://openalex.org/W3212023986","https://openalex.org/W4200631806","https://openalex.org/W4221143046","https://openalex.org/W4247091224","https://openalex.org/W4281657280","https://openalex.org/W4285077966","https://openalex.org/W4290927906","https://openalex.org/W4310419543","https://openalex.org/W4385567900","https://openalex.org/W4385570172","https://openalex.org/W4388994359","https://openalex.org/W4393147839","https://openalex.org/W4394743141","https://openalex.org/W4396723256","https://openalex.org/W4401834466","https://openalex.org/W4401856724","https://openalex.org/W4401862830","https://openalex.org/W4402671514","https://openalex.org/W6600351811","https://openalex.org/W6602641848","https://openalex.org/W6839328737","https://openalex.org/W6893809404"],"related_works":["https://openalex.org/W2373300491","https://openalex.org/W2395294869","https://openalex.org/W2378744544","https://openalex.org/W2594301978","https://openalex.org/W2379704676","https://openalex.org/W1998810860","https://openalex.org/W4206442282","https://openalex.org/W2384505857","https://openalex.org/W2355171581","https://openalex.org/W3204019825"],"abstract_inverted_index":{"We":[0],"focus":[1],"on":[2,58,151],"unlearning":[3,114],"unwanted":[4],"knowledge":[5,33,48],"in":[6,86,162],"autoregressive":[7],"large":[8],"language":[9],"models":[10],"(LLMs)":[11],"through":[12],"pruning.":[13],"Our":[14,132],"goal":[15],"is":[16],"to":[17,49,104,108,127,155],"selectively":[18,74],"remove":[19],"undesirable":[20,82,144],"information":[21],"(e.g.,":[22,34],"harmful":[23],"responses,":[24],"privacy-sensitive":[25],"data)":[26],"while":[27],"ensuring":[28],"the":[29,55,93,109,113,129,147,156],"preservation":[30],"of":[31,170],"desirable":[32,163],"positive":[35],"responses":[36],"and":[37,76,90,100,121],"objective":[38],"facts).":[39],"Previous":[40],"approaches":[41],"use":[42],"gradient":[43],"ascent":[44],"(GA)":[45],"over":[46],"undesired":[47,110],"inversely":[50],"optimize":[51],"LLMs,":[52],"which":[53],"compromises":[54],"model's":[56,130,148],"performance":[57,150,168],"desired":[59],"knowledge.":[60,83,111],"To":[61],"address":[62],"this":[63],"limitation,":[64],"we":[65,96,116],"introduce":[66],"a":[67,123],"novel":[68],"two-stage":[69],"approach,":[70],"named":[71],"LLM-Eraser,":[72],"for":[73],"identifying":[75],"editing":[77],"parameters":[78,106,120],"specifically":[79],"associated":[80],"with":[81,165],"LLM-Eraser":[84,141],"operates":[85],"two":[87],"stages:":[88],"localization":[89,94],"unlearning.":[91],"During":[92],"stage,":[95,115],"utilize":[97],"neuron":[98],"scores":[99],"trainable":[101],"soft":[102],"masks":[103],"identify":[105],"crucial":[107],"In":[112],"prune":[117],"these":[118],"identified":[119],"apply":[122],"selective":[124],"post-training":[125],"process":[126],"enhance":[128],"selectiveness.":[131],"experiments,":[133],"conducted":[134],"across":[135],"five":[136],"task":[137],"datasets,":[138],"demonstrate":[139],"that":[140],"effectively":[142],"unlearns":[143],"knowledge-evidenced":[145],"by":[146],"near-random":[149],"multiple-choice":[152],"questions":[153],"related":[154],"erased":[157],"knowledge-while":[158],"maintaining":[159],"high":[160],"proficiency":[161],"knowledge,":[164],"an":[166],"average":[167],"deficit":[169],"only":[171],"2.5%.":[172]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":1}],"updated_date":"2026-02-26T08:16:20.718346","created_date":"2025-10-10T00:00:00"}