{"id":"https://openalex.org/W7158277000","doi":"https://doi.org/10.48550/arxiv.2604.25642","title":"Prefill-Time Intervention for Mitigating Hallucination in Large Vision-Language Models","display_name":"Prefill-Time Intervention for Mitigating Hallucination in Large Vision-Language Models","publication_year":2026,"publication_date":"2026-04-28","ids":{"openalex":"https://openalex.org/W7158277000","doi":"https://doi.org/10.48550/arxiv.2604.25642"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.25642","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.25642","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.25642","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5123443205","display_name":"Chengsheng Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Chengsheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134845436","display_name":"Chenghao Sun","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sun, Chenghao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129511913","display_name":"Xinyan Jiang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiang, Xinyan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134845815","display_name":"Wei Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Wei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5134871570","display_name":"Xinmei Tian","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tian, Xinmei","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":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.2004999965429306,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.2004999965429306,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.08889999985694885,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.052000001072883606,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/generalizability-theory","display_name":"Generalizability theory","score":0.6432999968528748},{"id":"https://openalex.org/keywords/intervention","display_name":"Intervention (counseling)","score":0.6290000081062317},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.5245000123977661},{"id":"https://openalex.org/keywords/boosting","display_name":"Boosting (machine learning)","score":0.46619999408721924},{"id":"https://openalex.org/keywords/residual","display_name":"Residual","score":0.46070000529289246},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.3986000120639801},{"id":"https://openalex.org/keywords/filter","display_name":"Filter (signal processing)","score":0.3691999912261963}],"concepts":[{"id":"https://openalex.org/C27158222","wikidata":"https://www.wikidata.org/wiki/Q5532422","display_name":"Generalizability theory","level":2,"score":0.6432999968528748},{"id":"https://openalex.org/C2780665704","wikidata":"https://www.wikidata.org/wiki/Q959298","display_name":"Intervention (counseling)","level":2,"score":0.6290000081062317},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5852000117301941},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.5525000095367432},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.5245000123977661},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.5152999758720398},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.46619999408721924},{"id":"https://openalex.org/C155512373","wikidata":"https://www.wikidata.org/wiki/Q287450","display_name":"Residual","level":2,"score":0.46070000529289246},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.3986000120639801},{"id":"https://openalex.org/C106131492","wikidata":"https://www.wikidata.org/wiki/Q3072260","display_name":"Filter (signal processing)","level":2,"score":0.3691999912261963},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3612000048160553},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.3528999984264374},{"id":"https://openalex.org/C169345407","wikidata":"https://www.wikidata.org/wiki/Q8216221","display_name":"Uncorrelated","level":2,"score":0.3481000065803528},{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.3301999866962433},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.28690001368522644},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2842000126838684},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.2777999937534332},{"id":"https://openalex.org/C103088060","wikidata":"https://www.wikidata.org/wiki/Q1062839","display_name":"Error detection and correction","level":2,"score":0.25270000100135803}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.25642","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.25642","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.25642","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.25642","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"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],"progress":[7],"in":[8,36,142],"visual-textual":[9],"understanding,":[10],"yet":[11],"their":[12,55,134],"reliability":[13],"is":[14,104,116,157,171],"critically":[15],"undermined":[16],"by":[17],"hallucinations,":[18,38],"i.e.,":[19],"the":[20,46,59,89,93],"generation":[21],"of":[22,48],"factually":[23],"incorrect":[24],"or":[25],"inconsistent":[26],"responses.":[27],"While":[28],"recent":[29],"studies":[30],"using":[31],"steering":[32,82],"vectors":[33],"demonstrated":[34],"promise":[35],"reducing":[37],"a":[39,80],"notable":[40],"challenge":[41],"remains:":[42],"they":[43],"inadvertently":[44],"amplify":[45],"severity":[47],"residual":[49],"hallucinations.":[50],"We":[51],"attribute":[52],"this":[53],"to":[54,118,126,159],"exclusive":[56],"focus":[57],"on":[58],"decoding":[60,150],"stage,":[61,91],"where":[62],"errors":[63],"accumulate":[64],"autoregressively":[65],"and":[66,111,124,145,153,166],"progressively":[67],"worsen":[68],"subsequent":[69],"hallucinatory":[70],"outputs.":[71],"To":[72],"address":[73],"this,":[74],"we":[75],"propose":[76],"Prefill-Time":[77],"Intervention":[78],"(PTI),":[79],"novel":[81],"paradigm":[83],"that":[84],"intervenes":[85],"only":[86],"once":[87],"during":[88],"prefill":[90],"enhancing":[92],"initial":[94],"Key-Value":[95],"(KV)":[96],"cache":[97],"before":[98],"error":[99],"accumulation":[100],"occurs.":[101],"Specifically,":[102],"PTI":[103,156],"modality-aware,":[105],"deriving":[106],"distinct":[107],"directions":[108],"for":[109],"visual":[110],"textual":[112],"representations.":[113],"This":[114],"intervention":[115],"decoupled":[117],"steer":[119],"keys":[120],"toward":[121],"visually-grounded":[122],"objects":[123],"values":[125],"filter":[127],"background":[128],"noise,":[129],"correcting":[130],"hallucination-prone":[131],"representations":[132],"at":[133],"source.":[135],"Extensive":[136],"experiments":[137],"demonstrate":[138],"PTI's":[139],"significant":[140],"performance":[141],"mitigating":[143],"hallucinations":[144],"its":[146],"generalizability":[147],"across":[148],"diverse":[149],"strategies,":[151],"LVLMs,":[152],"benchmarks.":[154],"Moreover,":[155],"orthogonal":[158],"existing":[160],"decoding-stage":[161],"methods,":[162],"enabling":[163],"plug-and-play":[164],"integration":[165],"further":[167],"boosting":[168],"performance.":[169],"Code":[170],"available":[172],"at:":[173],"https://github.com/huaiyi66/PTI.":[174]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-04-30T00:00:00"}