{"id":"https://openalex.org/W4412945252","doi":"https://doi.org/10.18653/v1/2025.acl-long.634","title":"Activation Steering Decoding: Mitigating Hallucination in Large Vision-Language Models through Bidirectional Hidden State Intervention","display_name":"Activation Steering Decoding: Mitigating Hallucination in Large Vision-Language Models through Bidirectional Hidden State Intervention","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412945252","doi":"https://doi.org/10.18653/v1/2025.acl-long.634"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.acl-long.634","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.634","pdf_url":"https://aclanthology.org/2025.acl-long.634.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":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.acl-long.634.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5072056125","display_name":"Jingran Su","orcid":"https://orcid.org/0000-0002-9873-1770"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jingran Su","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101642745","display_name":"Jingfan Chen","orcid":"https://orcid.org/0000-0002-7559-6924"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jingfan Chen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100708466","display_name":"Hongxin Li","orcid":"https://orcid.org/0000-0002-2356-1607"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hongxin Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102821179","display_name":"Yuntao Chen","orcid":"https://orcid.org/0000-0002-9555-1897"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuntao Chen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100404155","display_name":"Qing Li","orcid":"https://orcid.org/0000-0003-0297-4346"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li Qing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5101589395","display_name":"Zhaoxiang Zhang","orcid":"https://orcid.org/0009-0002-5344-7159"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhaoxiang Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":6,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":2.1339,"has_fulltext":true,"cited_by_count":2,"citation_normalized_percentile":{"value":0.87665963,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":91,"max":98},"biblio":{"volume":null,"issue":null,"first_page":"12964","last_page":"12974"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10429","display_name":"EEG and Brain-Computer Interfaces","score":0.9836000204086304,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},"topics":[{"id":"https://openalex.org/T10429","display_name":"EEG and Brain-Computer Interfaces","score":0.9836000204086304,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.7447928190231323},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.6279975175857544},{"id":"https://openalex.org/keywords/intervention","display_name":"Intervention (counseling)","score":0.5500260591506958},{"id":"https://openalex.org/keywords/state","display_name":"State (computer science)","score":0.5186895132064819},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.41794317960739136},{"id":"https://openalex.org/keywords/speech-recognition","display_name":"Speech recognition","score":0.40747615694999695},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.34114402532577515},{"id":"https://openalex.org/keywords/computer-vision","display_name":"Computer vision","score":0.3310137391090393},{"id":"https://openalex.org/keywords/psychology","display_name":"Psychology","score":0.2602357864379883},{"id":"https://openalex.org/keywords/telecommunications","display_name":"Telecommunications","score":0.1803041696548462},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.12227088212966919}],"concepts":[{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.7447928190231323},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6279975175857544},{"id":"https://openalex.org/C2780665704","wikidata":"https://www.wikidata.org/wiki/Q959298","display_name":"Intervention (counseling)","level":2,"score":0.5500260591506958},{"id":"https://openalex.org/C48103436","wikidata":"https://www.wikidata.org/wiki/Q599031","display_name":"State (computer science)","level":2,"score":0.5186895132064819},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.41794317960739136},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.40747615694999695},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.34114402532577515},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.3310137391090393},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.2602357864379883},{"id":"https://openalex.org/C76155785","wikidata":"https://www.wikidata.org/wiki/Q418","display_name":"Telecommunications","level":1,"score":0.1803041696548462},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.12227088212966919},{"id":"https://openalex.org/C118552586","wikidata":"https://www.wikidata.org/wiki/Q7867","display_name":"Psychiatry","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.acl-long.634","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.634","pdf_url":"https://aclanthology.org/2025.acl-long.634.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":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.acl-long.634","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.634","pdf_url":"https://aclanthology.org/2025.acl-long.634.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":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/10","display_name":"Reduced inequalities","score":0.550000011920929}],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412945252.pdf","grobid_xml":"https://content.openalex.org/works/W4412945252.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W2161474341","https://openalex.org/W4302615923","https://openalex.org/W3203142394","https://openalex.org/W2351061015","https://openalex.org/W4220731478","https://openalex.org/W1974101135","https://openalex.org/W2017509870","https://openalex.org/W4251141768","https://openalex.org/W4360952157","https://openalex.org/W2153647085"],"abstract_inverted_index":{"Large":[0],"Vision-Language":[1],"Models":[2],"(LVLMs)":[3],"have":[4],"demonstrated":[5],"impressive":[6],"capabilities":[7],"in":[8,52,76,89],"multimodal":[9],"understanding,":[10],"but":[11],"they":[12],"frequently":[13],"suffer":[14],"from":[15],"hallucination":[16,31,72,88,117,132],"-generating":[17],"content":[18,46],"inconsistent":[19],"with":[20],"visual":[21,141],"inputs.In":[22],"this":[23,59],"work,":[24],"we":[25,61],"explore":[26],"a":[27,67,94,100],"novel":[28],"perspective":[29],"on":[30,139],"mitigation":[32],"by":[33,58,82],"examining":[34],"the":[35,53,77,90,106,121],"intermediate":[36,79],"activations":[37],"of":[38,87],"LVLMs":[39],"during":[40],"generation.Our":[41],"investigation":[42],"reveals":[43],"that":[44,70,104,127],"hallucinated":[45],"manifests":[47],"as":[48],"distinct,":[49],"identifiable":[50],"patterns":[51,86,118],"model's":[54,78,122],"hidden":[55],"state":[56],"space.Motivated":[57],"finding,":[60],"propose":[62],"Activation":[63],"Steering":[64],"Decoding":[65],"(ASD),":[66],"training-free":[68],"approach":[69,114,145],"mitigates":[71],"through":[73],"targeted":[74],"intervention":[75],"activations.ASD":[80],"operates":[81],"first":[83],"identifying":[84],"directional":[85],"activation":[91],"space":[92],"using":[93],"small":[95],"calibration":[96],"set,":[97],"then":[98],"employing":[99],"contrast":[101],"decoding":[102],"mechanism":[103],"computes":[105],"difference":[107],"between":[108],"positive":[109],"and":[110],"negative":[111],"steering":[112],"predictions.This":[113],"effectively":[115],"suppresses":[116],"while":[119,136],"preserving":[120],"general":[123,140],"capabilities.Extensive":[124],"experiments":[125],"demonstrate":[126],"our":[128,144],"method":[129],"significantly":[130],"reduces":[131],"across":[133],"multiple":[134],"benchmarks":[135],"maintaining":[137],"performance":[138],"understanding":[142],"tasks.Notably,":[143],"requires":[146],"no":[147],"model":[148],"re-training":[149],"or":[150],"architectural":[151],"modifications,":[152],"making":[153],"it":[154],"readily":[155],"applicable":[156],"to":[157],"existing":[158],"deployed":[159],"models.":[160]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":1}],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2025-10-10T00:00:00"}