{"id":"https://openalex.org/W4416034736","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.572","title":"Steering LVLMs via Sparse Autoencoder for Hallucination Mitigation","display_name":"Steering LVLMs via Sparse Autoencoder for Hallucination Mitigation","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416034736","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.572"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.572","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.572","pdf_url":"https://aclanthology.org/2025.findings-emnlp.572.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: EMNLP 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-emnlp.572.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101792313","display_name":"Zhengli Hua","orcid":"https://orcid.org/0000-0001-5789-7283"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhenglin Hua","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114654047","display_name":"Jinghan He","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jinghan He","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5040604135","display_name":"Zijun Yao","orcid":"https://orcid.org/0000-0003-3647-8770"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zijun Yao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Tianxu Han","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tianxu Han","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5085707125","display_name":"Haiyun Guo","orcid":"https://orcid.org/0000-0001-9241-6211"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Haiyun Guo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5013880628","display_name":"Yuheng Jia","orcid":"https://orcid.org/0000-0002-3907-6550"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuheng Jia","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5041233180","display_name":"Junfeng Fang","orcid":"https://orcid.org/0000-0003-2094-8678"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Junfeng Fang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"10808","last_page":"10828"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12553","display_name":"Psychedelics and Drug Studies","score":0.07989999651908875,"subfield":{"id":"https://openalex.org/subfields/3203","display_name":"Clinical Psychology"},"field":{"id":"https://openalex.org/fields/32","display_name":"Psychology"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T12553","display_name":"Psychedelics and Drug Studies","score":0.07989999651908875,"subfield":{"id":"https://openalex.org/subfields/3203","display_name":"Clinical Psychology"},"field":{"id":"https://openalex.org/fields/32","display_name":"Psychology"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.06909999996423721,"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/T12536","display_name":"Topological and Geometric Data Analysis","score":0.0357000008225441,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/autoencoder","display_name":"Autoencoder","score":0.5733000040054321},{"id":"https://openalex.org/keywords/pattern-recognition","display_name":"Pattern recognition (psychology)","score":0.45669999718666077},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.3750999867916107},{"id":"https://openalex.org/keywords/noise","display_name":"Noise (video)","score":0.30649998784065247},{"id":"https://openalex.org/keywords/sparse-approximation","display_name":"Sparse approximation","score":0.2639999985694885}],"concepts":[{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.628000020980835},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.593500018119812},{"id":"https://openalex.org/C101738243","wikidata":"https://www.wikidata.org/wiki/Q786435","display_name":"Autoencoder","level":3,"score":0.5733000040054321},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.45669999718666077},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.4302000105381012},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.3750999867916107},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.30649998784065247},{"id":"https://openalex.org/C124066611","wikidata":"https://www.wikidata.org/wiki/Q28684319","display_name":"Sparse approximation","level":2,"score":0.2639999985694885},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.2533999979496002},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.23119999468326569}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-emnlp.572","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.572","pdf_url":"https://aclanthology.org/2025.findings-emnlp.572.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: EMNLP 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.572","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.572","pdf_url":"https://aclanthology.org/2025.findings-emnlp.572.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: EMNLP 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G6467542141","display_name":null,"funder_award_id":"U24A20322","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416034736.pdf","grobid_xml":"https://content.openalex.org/works/W4416034736.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"vision-language":[1],"models":[2],"(LVLMs)":[3],"have":[4],"achieved":[5],"remarkable":[6],"performance":[7],"on":[8,34,130,146],"multimodal":[9],"tasks.However,":[10],"they":[11],"still":[12],"suffer":[13],"from":[14],"hallucinations,":[15,120,166],"generating":[16],"text":[17],"inconsistent":[18],"with":[19,98,174],"visual":[20],"input,":[21],"posing":[22],"significant":[23],"risks":[24],"in":[25,153,164],"real-world":[26],"applications.Existing":[27],"approaches":[28,163],"to":[29,55,71,77,92,150],"address":[30],"this":[31,85],"issue":[32],"focus":[33],"incorporating":[35],"external":[36],"knowledge":[37],"bases,":[38],"alignment":[39],"training,":[40],"or":[41,75,100],"decoding":[42,162],"strategies,":[43],"all":[44],"of":[45],"which":[46],"require":[47],"substantial":[48],"computational":[49],"cost":[50],"and":[51,105],"time.Recent":[52],"works":[53],"try":[54],"explore":[56],"more":[57,103],"efficient":[58],"alternatives":[59],"by":[60],"adjusting":[61],"LVLMs'":[62],"internal":[63],"representations.Although":[64],"promising,":[65],"these":[66,131],"methods":[67],"may":[68],"cause":[69],"hallucinations":[70,152],"be":[72],"insufficiently":[73],"suppressed":[74],"lead":[76],"excessive":[78],"interventions":[79,112],"that":[80,111,157],"negatively":[81],"affect":[82],"normal":[83],"semantics.In":[84],"work,":[86],"we":[87,133],"leverage":[88],"sparse":[89],"autoencoders":[90],"(SAEs)":[91],"identify":[93],"semantic":[94],"directions":[95,149],"closely":[96],"associated":[97],"faithfulness":[99],"hallucination,":[101],"extracting":[102],"precise":[104],"disentangled":[106],"hallucination-related":[107],"representations.Our":[108],"analysis":[109],"demonstrates":[110],"along":[113,123],"the":[114,124],"identified":[115],"faithful":[116],"direction":[117,126],"can":[118,127],"mitigate":[119,151],"while":[121,167],"those":[122],"hallucinatory":[125],"exacerbate":[128],"them.Building":[129],"insights,":[132],"propose":[134],"Steering":[135],"LVLMs":[136],"via":[137],"SAE":[138],"Latent":[139],"Directions":[140],"(SSL),":[141],"a":[142],"plug-and-play":[143],"method":[144],"based":[145],"SAE-derived":[147],"latent":[148],"LVLMs.Extensive":[154],"experiments":[155],"demonstrate":[156],"SSL":[158],"significantly":[159],"outperforms":[160],"existing":[161],"mitigating":[165],"maintaining":[168],"transferability":[169],"across":[170],"different":[171],"model":[172],"architectures":[173],"negligible":[175],"additional":[176],"time":[177],"overhead.The":[178],"code":[179],"is":[180],"available":[181],"at":[182],"https://github.com/huazhenglin2003/SSL.":[183]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-11-08T00:00:00"}