{"id":"https://openalex.org/W4412888393","doi":"https://doi.org/10.18653/v1/2025.findings-acl.589","title":"FairSteer: Inference Time Debiasing for LLMs with Dynamic Activation Steering","display_name":"FairSteer: Inference Time Debiasing for LLMs with Dynamic Activation Steering","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412888393","doi":"https://doi.org/10.18653/v1/2025.findings-acl.589"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.589","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.589","pdf_url":"https://aclanthology.org/2025.findings-acl.589.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 2025","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-acl.589.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5067453826","display_name":"Yichen Li","orcid":"https://orcid.org/0000-0002-0924-2466"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yichen Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Zhiting Fan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhiting Fan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103263792","display_name":"Ruizhe Chen","orcid":"https://orcid.org/0000-0001-5537-0082"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ruizhe Chen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111005062","display_name":"Xiaotang Gai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiaotang Gai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5084303078","display_name":"Luqi Gong","orcid":"https://orcid.org/0009-0000-8744-8630"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Luqi Gong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100456214","display_name":"Yan Zhang","orcid":"https://orcid.org/0000-0002-2267-8408"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yan Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5024343415","display_name":"Zuozhu Liu","orcid":"https://orcid.org/0000-0002-7816-502X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zuozhu Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":7,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":1.7588,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":{"value":0.87765021,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":91,"max":95},"biblio":{"volume":null,"issue":null,"first_page":"11293","last_page":"11312"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12761","display_name":"Data Stream Mining Techniques","score":0.9495000243186951,"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/T12761","display_name":"Data Stream Mining Techniques","score":0.9495000243186951,"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/T11182","display_name":"Auction Theory and Applications","score":0.9351000189781189,"subfield":{"id":"https://openalex.org/subfields/1803","display_name":"Management Science and Operations Research"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/debiasing","display_name":"Debiasing","score":0.962471604347229},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6094757914543152},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.589677095413208},{"id":"https://openalex.org/keywords/psychology","display_name":"Psychology","score":0.18083307147026062},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.17375648021697998},{"id":"https://openalex.org/keywords/cognitive-science","display_name":"Cognitive science","score":0.13548126816749573}],"concepts":[{"id":"https://openalex.org/C2779458634","wikidata":"https://www.wikidata.org/wiki/Q24963715","display_name":"Debiasing","level":2,"score":0.962471604347229},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6094757914543152},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.589677095413208},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.18083307147026062},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.17375648021697998},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.13548126816749573}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.589","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.589","pdf_url":"https://aclanthology.org/2025.findings-acl.589.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 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-acl.589","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.589","pdf_url":"https://aclanthology.org/2025.findings-acl.589.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 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G4474290519","display_name":null,"funder_award_id":"LZ23F020008","funder_id":"https://openalex.org/F4320338464","funder_display_name":"Natural Science Foundation of Zhejiang Province"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"},{"id":"https://openalex.org/F4320322927","display_name":"Zhejiang University","ror":"https://ror.org/00a2xv884"},{"id":"https://openalex.org/F4320338464","display_name":"Natural Science Foundation of Zhejiang Province","ror":"https://ror.org/01h0zpd94"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412888393.pdf","grobid_xml":"https://content.openalex.org/works/W4412888393.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W4362554880","https://openalex.org/W4281684980","https://openalex.org/W4386875279","https://openalex.org/W2171721708","https://openalex.org/W4390963114","https://openalex.org/W4287887864","https://openalex.org/W4388144300"],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2],"(LLMs)":[3],"are":[4],"prone":[5],"to":[6,13,24,27,104],"capturing":[7],"biases":[8],"from":[9,118],"training":[10],"corpus,":[11],"leading":[12],"potential":[14],"negative":[15],"social":[16],"impacts.Existing":[17],"prompt-based":[18],"debiasing":[19,48,88,125],"methods":[20],"exhibit":[21],"instability":[22],"due":[23],"their":[25],"sensitivity":[26],"prompt":[28,53,121],"changes,":[29],"while":[30],"fine-tuning-based":[31],"techniques":[32],"incur":[33],"substantial":[34],"computational":[35],"overhead":[36],"and":[37,93,110,149],"catastrophic":[38],"forgetting.In":[39],"this":[40],"paper,":[41],"we":[42],"propose":[43],"FairSteer,":[44],"a":[45,100],"novel":[46],"inference-time":[47],"framework":[49],"without":[50],"requiring":[51],"customized":[52],"design":[54],"or":[55],"model":[56],"retraining.Motivated":[57],"by":[58,126],"the":[59,77,132,140],"linear":[60,102],"representation":[61],"hypothesis,":[62],"our":[63],"preliminary":[64],"investigation":[65],"demonstrates":[66,139],"that":[67],"fairness-related":[68],"features":[69],"can":[70],"be":[71],"encoded":[72],"into":[73],"separable":[74],"directions":[75,116],"in":[76,82,108,131],"hidden":[78],"activation":[79,86,95],"space.FairSteer":[80],"operates":[81],"three":[83],"steps:":[84],"biased":[85],"detection,":[87],"steering":[89],"vector":[90],"(DSV)":[91],"computation,":[92],"dynamic":[94],"steering.Specifically,":[96],"it":[97,123],"first":[98],"trains":[99],"lightweight":[101],"classifier":[103],"detect":[105],"bias":[106],"signatures":[107],"activations,":[109],"then":[111],"computes":[112],"DSVs":[113,130],"as":[114],"intervention":[115],"derived":[117],"small":[119],"contrastive":[120],"pairs.Subsequently,":[122],"performs":[124],"adjusting":[127],"activations":[128],"with":[129,136],"inference":[133],"stage.Comprehensive":[134],"evaluation":[135,148],"six":[137],"LLMs":[138],"superiority":[141],"of":[142],"FairSteer":[143],"across":[144],"question-answering,":[145],"counterfactual":[146],"input":[147],"open-ended":[150],"text":[151],"generation":[152],"tasks.":[153]},"counts_by_year":[{"year":2025,"cited_by_count":1}],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2025-10-10T00:00:00"}