{"id":"https://openalex.org/W4417209574","doi":"https://doi.org/10.1609/aaai.v40i44.41080","title":"Uncovering and Aligning Anomalous Attention Heads to Defend Against NLP Backdoor Attacks","display_name":"Uncovering and Aligning Anomalous Attention Heads to Defend Against NLP Backdoor Attacks","publication_year":2026,"publication_date":"2026-03-14","ids":{"openalex":"https://openalex.org/W4417209574","doi":"https://doi.org/10.1609/aaai.v40i44.41080"},"language":"en","primary_location":{"id":"doi:10.1609/aaai.v40i44.41080","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i44.41080","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/41080/45041","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://ojs.aaai.org/index.php/AAAI/article/download/41080/45041","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Haotian Jin","orcid":null},"institutions":[{"id":"https://openalex.org/I4210156404","display_name":"Institute of Information Engineering","ror":"https://ror.org/04r53se39","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210156404"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Haotian Jin","raw_affiliation_strings":["Institute of Information Engineering, Chinese Academy of Sciences\nState Key Laboratory of Cyberspace Security Defense\nSchool of Cyber Security, University of Chinese Academy of Sciences"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Information Engineering, Chinese Academy of Sciences\nState Key Laboratory of Cyberspace Security Defense\nSchool of Cyber Security, University of Chinese Academy of Sciences","institution_ids":["https://openalex.org/I4210156404"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5115597085","display_name":"Yang Li","orcid":"https://orcid.org/0009-0005-3334-4453"},"institutions":[{"id":"https://openalex.org/I4210156404","display_name":"Institute of Information Engineering","ror":"https://ror.org/04r53se39","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210156404"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yang Li","raw_affiliation_strings":["Institute of Information Engineering, Chinese Academy of Sciences\nState Key Laboratory of Cyberspace Security Defense"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Information Engineering, Chinese Academy of Sciences\nState Key Laboratory of Cyberspace Security Defense","institution_ids":["https://openalex.org/I4210156404"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5040867646","display_name":"Haihui Fan","orcid":"https://orcid.org/0000-0002-4366-9121"},"institutions":[{"id":"https://openalex.org/I4210156404","display_name":"Institute of Information Engineering","ror":"https://ror.org/04r53se39","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210156404"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Haihui Fan","raw_affiliation_strings":["Institute of Information Engineering, Chinese Academy of Sciences\nState Key Laboratory of Cyberspace Security Defense"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Information Engineering, Chinese Academy of Sciences\nState Key Laboratory of Cyberspace Security Defense","institution_ids":["https://openalex.org/I4210156404"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100984926","display_name":"Lin Shen","orcid":"https://orcid.org/0000-0001-6304-6223"},"institutions":[{"id":"https://openalex.org/I4210156404","display_name":"Institute of Information Engineering","ror":"https://ror.org/04r53se39","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210156404"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Lin Shen","raw_affiliation_strings":["Institute of Information Engineering, Chinese Academy of Sciences\nState Key Laboratory of Cyberspace Security Defense\nSchool of Cyber Security, University of Chinese Academy of Sciences"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Information Engineering, Chinese Academy of Sciences\nState Key Laboratory of Cyberspace Security Defense\nSchool of Cyber Security, University of Chinese Academy of Sciences","institution_ids":["https://openalex.org/I4210156404"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5063982441","display_name":"Xiangfang Li","orcid":"https://orcid.org/0000-0003-0334-4733"},"institutions":[{"id":"https://openalex.org/I4210156404","display_name":"Institute of Information Engineering","ror":"https://ror.org/04r53se39","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210156404"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xiangfang Li","raw_affiliation_strings":["Institute of Information Engineering, Chinese Academy of Sciences\nState Key Laboratory of Cyberspace Security Defense\nSchool of Cyber Security, University of Chinese Academy of Sciences"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Information Engineering, Chinese Academy of Sciences\nState Key Laboratory of Cyberspace Security Defense\nSchool of Cyber Security, University of Chinese Academy of Sciences","institution_ids":["https://openalex.org/I4210156404"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5100732385","display_name":"Bo Li","orcid":"https://orcid.org/0000-0001-7756-8030"},"institutions":[{"id":"https://openalex.org/I4210156404","display_name":"Institute of Information Engineering","ror":"https://ror.org/04r53se39","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210156404"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Bo Li","raw_affiliation_strings":["Institute of Information Engineering, Chinese Academy of Sciences\nState Key Laboratory of Cyberspace Security Defense"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Information Engineering, Chinese Academy of Sciences\nState Key Laboratory of Cyberspace Security Defense","institution_ids":["https://openalex.org/I4210156404"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I4210156404"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":"40","issue":"44","first_page":"37472","last_page":"37480"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.8730999827384949,"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":0.8730999827384949,"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/T11424","display_name":"Security and Verification in Computing","score":0.02669999934732914,"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/T11241","display_name":"Advanced Malware Detection Techniques","score":0.02160000056028366,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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/backdoor","display_name":"Backdoor","score":0.9995999932289124},{"id":"https://openalex.org/keywords/flexibility","display_name":"Flexibility (engineering)","score":0.6065999865531921},{"id":"https://openalex.org/keywords/similarity","display_name":"Similarity (geometry)","score":0.2282000035047531},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.22689999639987946},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.21050000190734863}],"concepts":[{"id":"https://openalex.org/C2781045450","wikidata":"https://www.wikidata.org/wiki/Q254569","display_name":"Backdoor","level":2,"score":0.9995999932289124},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6646999716758728},{"id":"https://openalex.org/C2780598303","wikidata":"https://www.wikidata.org/wiki/Q65921492","display_name":"Flexibility (engineering)","level":2,"score":0.6065999865531921},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4512999951839447},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.38260000944137573},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3124000132083893},{"id":"https://openalex.org/C103278499","wikidata":"https://www.wikidata.org/wiki/Q254465","display_name":"Similarity (geometry)","level":3,"score":0.2282000035047531},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.22689999639987946},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.21050000190734863},{"id":"https://openalex.org/C59577422","wikidata":"https://www.wikidata.org/wiki/Q10265143","display_name":"False accusation","level":2,"score":0.20720000565052032}],"mesh":[],"locations_count":5,"locations":[{"id":"doi:10.1609/aaai.v40i44.41080","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i44.41080","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/41080/45041","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},{"id":"pmh:oai:arXiv.org:2511.13789","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2511.13789","pdf_url":"https://arxiv.org/pdf/2511.13789","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"pmh:oai:arXiv.org:2511.13789","is_oa":true,"landing_page_url":"https://arxiv.org/abs/2511.13789","pdf_url":"https://arxiv.org/pdf/2511.13789","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"pmh:oai:ojs.aaai.org:article/41080","is_oa":false,"landing_page_url":"https://ojs.aaai.org/index.php/AAAI/article/view/41080","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"2159-5399","raw_type":"info:eu-repo/semantics/article"},{"id":"doi:10.48550/arxiv.2511.13789","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2511.13789","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.1609/aaai.v40i44.41080","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i44.41080","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/41080/45041","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320321133","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4417209574.pdf","grobid_xml":"https://content.openalex.org/works/W4417209574.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Backdoor":[0],"attacks":[1,108,163],"pose":[2],"a":[3,83],"serious":[4],"threat":[5],"to":[6,16,34,50,63,106,118,135],"the":[7,98,144,158,166],"security":[8],"of":[9,26,66,97,146,161],"large":[10],"language":[11],"models":[12,104],"(LLMs),":[13],"causing":[14],"them":[15],"exhibit":[17,109],"anomalous":[18],"behavior":[19],"under":[20],"specific":[21,54,64],"trigger":[22,43],"conditions.":[23],"The":[24],"design":[25,44],"backdoor":[27,58,84,92,107,147,162],"triggers":[28,33,67],"has":[29],"evolved":[30],"from":[31],"fixed":[32],"dynamic":[35],"or":[36,68],"implicit":[37],"triggers.":[38,119],"This":[39],"increased":[40],"flexibility":[41],"in":[42],"makes":[45],"it":[46],"challenging":[47],"for":[48,75],"defenders":[49],"accurately":[51],"identify":[52],"their":[53],"forms.":[55],"Most":[56],"existing":[57],"defense":[59],"methods":[60],"are":[61],"limited":[62],"types":[65],"rely":[69],"on":[70,88,121,169],"an":[71,126],"additional":[72],"clean":[73],"model":[74],"support.":[76],"To":[77],"address":[78],"this":[79,122],"issue,":[80],"we":[81,124],"propose":[82,125],"detection":[85,93],"method":[86,155],"based":[87],"attention":[89,114,127,139],"similarity,":[90],"enabling":[91],"without":[94],"prior":[95],"knowledge":[96],"trigger.":[99],"Our":[100],"study":[101],"reveals":[102],"that":[103,153],"subjected":[105],"unusually":[110],"high":[111],"similarity":[112],"among":[113],"heads":[115],"when":[116],"exposed":[117],"Based":[120],"observation,":[123],"safety":[128],"alignment":[129],"approach":[130],"combined":[131],"with":[132],"head-wise":[133],"fine-tuning":[134],"rectify":[136],"potentially":[137],"contaminated":[138],"heads,":[140],"thereby":[141],"effectively":[142],"mitigating":[143],"impact":[145],"attacks.":[148],"Extensive":[149],"experimental":[150],"results":[151],"demonstrate":[152],"our":[154],"significantly":[156],"reduces":[157],"success":[159],"rate":[160],"while":[164],"preserving":[165],"model\u2019s":[167],"performance":[168],"downstream":[170],"tasks.":[171]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-11-20T00:00:00"}