{"id":"https://openalex.org/W7204158795","doi":"https://doi.org/10.18653/2026.emnlp-main.953","title":"BLADE: Bilevel Low-rank Augmented-Lagrangian Erasure for LLM Unlearning","display_name":"BLADE: Bilevel Low-rank Augmented-Lagrangian Erasure for LLM Unlearning","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7204158795","doi":"https://doi.org/10.18653/2026.emnlp-main.953"},"language":"en","primary_location":{"id":"doi:10.18653/2026.emnlp-main.953","doi":"https://doi.org/10.18653/2026.emnlp-main.953","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.emnlp-main.953","pdf_url":"https://aclanthology.org/2026.emnlp-main.953.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"bronze","oa_url":"https://aclanthology.org/2026.emnlp-main.953.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5148081042","display_name":"Md Toufikuzzaman","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I130769515","display_name":"Pennsylvania State University","ror":"https://ror.org/04p491231","country_code":"US","type":"education","lineage":["https://openalex.org/I130769515"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Md Toufikuzzaman","raw_affiliation_strings":["The Pennsylvania State University , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Pennsylvania State University , USA","institution_ids":["https://openalex.org/I130769515"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101774436","display_name":"Ahmad Mousavi","orcid":"https://orcid.org/0000-0003-4518-5857","observed_orcids":["https://orcid.org/0000-0003-4518-5857"]},"institutions":[{"id":"https://openalex.org/I181401687","display_name":"American University","ror":"https://ror.org/052w4zt36","country_code":"US","type":"education","lineage":["https://openalex.org/I181401687"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ahmad Mousavi","raw_affiliation_strings":["American University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"American University","institution_ids":["https://openalex.org/I181401687"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5148133968","display_name":"Dongwon Lee","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I130769515","display_name":"Pennsylvania State University","ror":"https://ror.org/04p491231","country_code":"US","type":"education","lineage":["https://openalex.org/I130769515"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Dongwon Lee","raw_affiliation_strings":["The Pennsylvania State University , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Pennsylvania State University , USA","institution_ids":["https://openalex.org/I130769515"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"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":0.67116985,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"21966","last_page":"21987"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.9951000213623047,"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/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.9951000213623047,"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":0.0008999999845400453,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.00039999998989515007,"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/llm-unlearning","display_name":"LLM unlearning","score":0.9319999814033508},{"id":"https://openalex.org/keywords/bi-level-optimization","display_name":"bi-level optimization","score":0.949999988079071},{"id":"https://openalex.org/keywords/low-rank-optimization","display_name":"low-rank optimization","score":0.6589999794960022},{"id":"https://openalex.org/keywords/augmented-lagrangian-method","display_name":"augmented Lagrangian method","score":0.9760000109672546},{"id":"https://openalex.org/keywords/lora-adapters","display_name":"LoRA adapters","score":0.7609999775886536},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.5009999871253967}],"concepts":[{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.7254999876022339},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6338000297546387},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.6047000288963318},{"id":"https://openalex.org/C3309286","wikidata":"https://www.wikidata.org/wiki/Q4907693","display_name":"Bilevel optimization","level":3,"score":0.5918999910354614},{"id":"https://openalex.org/C7149132","wikidata":"https://www.wikidata.org/wiki/Q1377840","display_name":"Forgetting","level":2,"score":0.5820000171661377},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.5382999777793884},{"id":"https://openalex.org/C2778790127","wikidata":"https://www.wikidata.org/wiki/Q484885","display_name":"Erasure","level":2,"score":0.5120999813079834},{"id":"https://openalex.org/C171268870","wikidata":"https://www.wikidata.org/wiki/Q1486676","display_name":"GRASP","level":2,"score":0.5087000131607056},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.48590001463890076},{"id":"https://openalex.org/C150452318","wikidata":"https://www.wikidata.org/wiki/Q4820432","display_name":"Augmented Lagrangian method","level":2,"score":0.46959999203681946},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.43459999561309814},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.42489999532699585},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.39100000262260437},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.3880999982357025},{"id":"https://openalex.org/C189645446","wikidata":"https://www.wikidata.org/wiki/Q350865","display_name":"Mirroring","level":2,"score":0.36390000581741333},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.30970001220703125},{"id":"https://openalex.org/C126042441","wikidata":"https://www.wikidata.org/wiki/Q1324888","display_name":"Frame (networking)","level":2,"score":0.2833999991416931},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2802000045776367},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2784999907016754},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.274399995803833},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.25380000472068787}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.18653/2026.emnlp-main.953","doi":"https://doi.org/10.18653/2026.emnlp-main.953","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.emnlp-main.953","pdf_url":"https://aclanthology.org/2026.emnlp-main.953.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2608.22557","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.22557","pdf_url":"https://arxiv.org/pdf/2608.22557","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","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2608.22557","doi":"https://doi.org/10.48550/arxiv.2608.22557","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.22557","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","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.18653/2026.emnlp-main.953","doi":"https://doi.org/10.18653/2026.emnlp-main.953","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.emnlp-main.953","pdf_url":"https://aclanthology.org/2026.emnlp-main.953.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G7615569247","display_name":"SFS AI + Cybersecurity: Artificial Intelligence (AI)-Integrated Training for Future Cybersecurity Workforce","funder_award_id":"2438810","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G9001309218","display_name":"Collaborative Research: SaTC 2.0: RES: Local Rules, Global Models: Jurisdiction- Aware Solutions for Copyright Vulnerabilities in LLMs","funder_award_id":"2555559","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7204158795.pdf","grobid_xml":"https://content.openalex.org/works/W7204158795.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7111250527","https://openalex.org/W7133324420","https://openalex.org/W6979253563","https://openalex.org/W7155042273","https://openalex.org/W7124358576","https://openalex.org/W7084397624","https://openalex.org/W4403573440","https://openalex.org/W4407091506","https://openalex.org/W4415798720","https://openalex.org/W4417536748"],"abstract_inverted_index":{"Existing":[0],"LLM":[1],"unlearning":[2,131],"methods":[3,23],"struggle":[4],"with":[5],"robustness:":[6],"unbounded":[7],"forget":[8,55],"losses":[9],"degrade":[10],"model":[11],"coherence,":[12],"fixed-weight":[13],"balancing":[14],"cannot":[15],"adapt":[16],"as":[17],"retain":[18,75,90],"difficulty":[19],"shifts":[20],"midtraining,":[21],"and":[22,80,117,121,128],"that":[24,72,88],"work":[25],"on":[26,111,114,119,133],"one":[27],"benchmark":[28,99],"falter":[29],"under":[30,125],"scaling":[31,127],"or":[32],"repeated":[33],"application.We":[34],"propose":[35],"BLADE,":[36],"1":[37],"a":[38,53,63,81],"constrained":[39],"bilevel":[40,82],"framework":[41],"whose":[42,57],"three":[43,98],"mechanisms":[44],"give":[45],"smooth,":[46],"predictable":[47],"control":[48],"over":[49,105],"the":[50,106,137],"optimization":[51],"landscape:":[52],"clamped-entropy":[54],"loss":[56],"gradient":[58],"is":[59],"exactly":[60],"zero":[61],"once":[62],"token":[64],"reaches":[65],"sufficient":[66],"uncertainty;":[67],"an":[68],"asymmetric":[69],"augmented":[70],"Lagrangian":[71],"permanently":[73],"ratchets":[74],"protection":[76],"after":[77],"any":[78],"violation;":[79],"structure":[83],"confined":[84],"to":[85],"LoRA":[86],"adapters":[87],"repairs":[89],"damage":[91],"before":[92],"each":[93],"forgetting":[94],"step.BLADE":[95],"dominates":[96],"across":[97],"families,":[100],"improving":[101],"average":[102],"composite":[103],"scores":[104],"strongest":[107],"baselines":[108],"by":[109],"6%":[110],"TOFU,":[112],"9%":[113],"MUSE":[115,134],"Books,":[116],"7%":[118],"KnowUndo,":[120],"it":[122],"remains":[123],"stable":[124],"4\u00d7":[126],"4":[129],"sequential":[130],"steps":[132],"News":[135],"where":[136],"best":[138],"competing":[139],"method":[140],"collapses":[141],"entirely.":[142]},"counts_by_year":[],"updated_date":"2026-11-03T10:50:19.383937","created_date":"2026-08-26T00:00:00"}