{"id":"https://openalex.org/W7166869350","doi":"https://doi.org/10.18653/v1/2026.findings-acl.815","title":"DiSec: Mitigating Backdoors in Pre-trained Language Models via Disentanglement of Adversarial Weights for Secure Fine-Tuning","display_name":"DiSec: Mitigating Backdoors in Pre-trained Language Models via Disentanglement of Adversarial Weights for Secure Fine-Tuning","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166869350","doi":"https://doi.org/10.18653/v1/2026.findings-acl.815"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-acl.815","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.815","pdf_url":"https://aclanthology.org/2026.findings-acl.815.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: ACL 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-acl.815.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5139813663","display_name":"Sunanda Das","orcid":null},"institutions":[{"id":"https://openalex.org/I78715868","display_name":"University of Arkansas at Fayetteville","ror":"https://ror.org/05jbt9m15","country_code":"US","type":"education","lineage":["https://openalex.org/I78715868"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Sunanda Das","raw_affiliation_strings":["Dept. of EECS , University of Arkansas , Fayetteville , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Dept. of EECS , University of Arkansas , Fayetteville , USA","institution_ids":["https://openalex.org/I78715868"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5139762118","display_name":"Qinghua Li","orcid":null},"institutions":[{"id":"https://openalex.org/I78715868","display_name":"University of Arkansas at Fayetteville","ror":"https://ror.org/05jbt9m15","country_code":"US","type":"education","lineage":["https://openalex.org/I78715868"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Qinghua Li","raw_affiliation_strings":["Dept. of EECS , University of Arkansas , Fayetteville , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Dept. of EECS , University of Arkansas , Fayetteville , USA","institution_ids":["https://openalex.org/I78715868"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I78715868"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.8581232,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"16540","last_page":"16559"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9768999814987183,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9768999814987183,"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.004000000189989805,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.0015999999595806003,"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/adversarial-system","display_name":"Adversarial system","score":0.5917999744415283},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.41920000314712524},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.34869998693466187},{"id":"https://openalex.org/keywords/identification","display_name":"Identification (biology)","score":0.33980000019073486},{"id":"https://openalex.org/keywords/threat-model","display_name":"Threat model","score":0.28540000319480896}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6996999979019165},{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.5917999744415283},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.41920000314712524},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3700999915599823},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.34869998693466187},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.33980000019073486},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.2892000079154968},{"id":"https://openalex.org/C140547941","wikidata":"https://www.wikidata.org/wiki/Q7797194","display_name":"Threat model","level":2,"score":0.28540000319480896},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.25940001010894775},{"id":"https://openalex.org/C2776036281","wikidata":"https://www.wikidata.org/wiki/Q48769818","display_name":"Constraint (computer-aided design)","level":2,"score":0.2590000033378601},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.2556999921798706},{"id":"https://openalex.org/C2779696439","wikidata":"https://www.wikidata.org/wiki/Q7512811","display_name":"Signature (topology)","level":2,"score":0.25540000200271606}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-acl.815","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.815","pdf_url":"https://aclanthology.org/2026.findings-acl.815.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: ACL 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-acl.815","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.815","pdf_url":"https://aclanthology.org/2026.findings-acl.815.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: ACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/10","display_name":"Reduced inequalities","score":0.7427970170974731}],"awards":[],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320316941","display_name":"Arkansas Economic Development Commission","ror":"https://ror.org/01b047235"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166869350.pdf","grobid_xml":"https://content.openalex.org/works/W7166869350.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Task-agnostic":[0],"backdoor":[1,133],"attacks":[2],"can":[3],"contaminate":[4],"pre-trained":[5],"language":[6],"models":[7],"(PLMs)":[8],"in":[9,47],"a":[10,52,109,145],"way":[11],"that":[12,58,86,121],"survives":[13],"downstream":[14,69,128],"adaptation,":[15],"even":[16],"under":[17],"full":[18],"fine-tuning,":[19],"making":[20],"it":[21],"difficult":[22],"for":[23],"practitioners":[24],"to":[25,37,81,148],"trust":[26],"third-party":[27],"checkpoints.Existing":[28],"defenses":[29],"often":[30],"rely":[31,67],"on":[32,68,124],"privileged":[33],"assumptions":[34],"(e.g.,":[35],"access":[36],"poisoned":[38],"data":[39,80],"or":[40,71],"trigger/target":[41],"knowledge),":[42],"thereby":[43],"limiting":[44],"their":[45],"applicability":[46],"realistic":[48],"settings.We":[49],"present":[50],"DiSec,":[51],"robust":[53],"and":[54,64,92,102,131],"label-efficient":[55],"purification":[56],"framework":[57],"uses":[59],"only":[60,123],"clean":[61,79],"auxiliary":[62],"text":[63],"does":[65],"not":[66],"supervision":[70],"attack":[72,138],"signatures.DiSec":[73],"elicits":[74],"model-internal":[75],"signals":[76],"from":[77],"this":[78],"separate":[82],"suspicious":[83],"parameter":[84],"components":[85],"are":[87,156],"inconsistent":[88],"with":[89],"benign":[90],"behavior,":[91],"then":[93],"flags":[94],"anomalous":[95],"structures":[96],"by":[97],"jointly":[98],"leveraging":[99],"complementary":[100],"spectral":[101],"generative":[103],"views":[104],"of":[105],"outliers.Finally,":[106],"DiSec":[107,135],"performs":[108],"structure-preserving":[110],"repair":[111],"via":[112],"layer-local":[113],"prototype-based":[114],"mean":[115],"correction,":[116],"yielding":[117],"an":[118],"idempotent":[119],"update":[120],"depends":[122],"nonadversarial":[125],"statistics.Across":[126],"diverse":[127],"classification":[129],"tasks":[130],"PLM":[132],"strategies,":[134],"substantially":[136],"suppresses":[137],"success":[139],"while":[140],"preserving":[141],"clean-task":[142],"utility,":[143],"offering":[144],"practical":[146],"path":[147],"securing":[149],"fully":[150],"finetuned":[151],"PLMs":[152],"before":[153],"deployment.The":[154],"codes":[155],"publicly":[157],"available":[158],"at":[159]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}