{"id":"https://openalex.org/W7164053023","doi":"https://doi.org/10.48550/arxiv.2606.07647","title":"Steer Where It Matters: Token-Level Visual-Sensitivity Steering for LVLMs Hallucination Mitigation","display_name":"Steer Where It Matters: Token-Level Visual-Sensitivity Steering for LVLMs Hallucination Mitigation","publication_year":2026,"publication_date":"2026-06-02","ids":{"openalex":"https://openalex.org/W7164053023","doi":"https://doi.org/10.48550/arxiv.2606.07647"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.07647","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.07647","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":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.2606.07647","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138213128","display_name":"Ruipeng Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Ruipeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138286119","display_name":"Zhihao Li (359073)","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Zhihao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138242278","display_name":"C. L. Philip Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, C. L. Philip","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5138220404","display_name":"Tong Zhang (103827)","orcid":"https://orcid.org/0009-0007-7343-4175"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Tong","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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.5016000270843506,"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":0.5016000270843506,"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":0.05550000071525574,"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/T12357","display_name":"Digital Media Forensic Detection","score":0.04969999939203262,"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/controllability","display_name":"Controllability","score":0.6909999847412109},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5037000179290771},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.43799999356269836},{"id":"https://openalex.org/keywords/sequence","display_name":"Sequence (biology)","score":0.4230000078678131},{"id":"https://openalex.org/keywords/autoregressive-model","display_name":"Autoregressive model","score":0.4223000109195709},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.4214000105857849},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.415800005197525},{"id":"https://openalex.org/keywords/control-theory","display_name":"Control theory (sociology)","score":0.40799999237060547}],"concepts":[{"id":"https://openalex.org/C48209547","wikidata":"https://www.wikidata.org/wiki/Q1331104","display_name":"Controllability","level":2,"score":0.6909999847412109},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6662999987602234},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5037000179290771},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.45509999990463257},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.43799999356269836},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.4230000078678131},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.4223000109195709},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.4214000105857849},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.415800005197525},{"id":"https://openalex.org/C47446073","wikidata":"https://www.wikidata.org/wiki/Q5165890","display_name":"Control theory (sociology)","level":3,"score":0.40799999237060547},{"id":"https://openalex.org/C89611455","wikidata":"https://www.wikidata.org/wiki/Q6804646","display_name":"Mechanism (biology)","level":2,"score":0.36880001425743103},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.3578999936580658},{"id":"https://openalex.org/C133731056","wikidata":"https://www.wikidata.org/wiki/Q4917288","display_name":"Control engineering","level":1,"score":0.34950000047683716},{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.32010000944137573},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.30979999899864197},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.3041999936103821},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.3012999892234802},{"id":"https://openalex.org/C127162648","wikidata":"https://www.wikidata.org/wiki/Q16858953","display_name":"Channel (broadcasting)","level":2,"score":0.29980000853538513},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.28700000047683716},{"id":"https://openalex.org/C48103436","wikidata":"https://www.wikidata.org/wiki/Q599031","display_name":"State (computer science)","level":2,"score":0.2775000035762787},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2612999975681305},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.2597000002861023}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.07647","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.07647","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:10.48550/arxiv.2606.07647","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.07647","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":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"vision":[1],"language":[2],"models":[3],"(LVLMs)":[4],"have":[5],"made":[6],"rapid":[7],"advancements":[8],"and":[9,31,49,54,93,117,120,140,177],"are":[10],"deployed":[11],"across":[12,51,144],"various":[13],"applications,":[14],"yet":[15],"hallucinations":[16],"remain":[17],"a":[18,81],"major":[19],"challenge.":[20],"Activation":[21],"steering":[22,74,83,115,125,151,184],"is":[23],"appealing":[24],"due":[25],"to":[26],"its":[27],"minimal":[28,136],"training":[29,137],"overhead":[30],"controllability":[32],"at":[33,153],"inference":[34],"time.":[35],"However,":[36],"we":[37,101],"found":[38],"that":[39,58],"during":[40],"autoregressive":[41],"decoding,":[42],"visual":[43],"conditioning":[44],"affects":[45],"token":[46],"prediction":[47],"sparsely":[48],"locally":[50],"decoding":[52,155],"steps,":[53],"many":[55,77],"existing":[56,78],"methods":[57,79],"average":[59],"image-versus-no-image":[60],"differences":[61],"over":[62,182],"the":[63,87,150],"entire":[64],"sequence":[65],"dilute":[66],"these":[67,99],"critical":[68],"signals,":[69],"yielding":[70],"low":[71],"signal-to-noise":[72],"ratio":[73],"directions.":[75],"Additionally,":[76],"apply":[80],"fixed":[82],"strength,":[84],"which":[85],"misallocates":[86],"intervention":[88],"budget,":[89],"over-perturbs":[90],"non-critical":[91],"tokens,":[92],"can":[94,141],"cause":[95],"instability.":[96],"To":[97],"address":[98],"limitations,":[100],"propose":[102],"Token-Level":[103],"Visual-Sensitivity":[104],"Steering":[105],"(TLVS)":[106],"for":[107,138],"hallucination":[108],"mitigation.":[109],"Our":[110],"approach":[111],"first":[112],"extracts":[113],"token-level":[114],"vectors":[116],"refines":[118],"them,":[119],"then":[121],"applies":[122],"fine-grained,":[123],"visual-sensitivity-adaptive":[124],"only":[126,135],"where":[127],"it":[128],"matters.":[129],"This":[130],"lightweight,":[131],"plug-and-play":[132],"mechanism":[133],"requires":[134],"calibration":[139],"be":[142],"applied":[143],"diverse":[145],"vision-language":[146],"models.":[147],"It":[148],"modulates":[149],"strength":[152],"each":[154],"step,":[156],"selectively":[157],"suppressing":[158],"hallucination-prone":[159],"spans":[160],"while":[161],"preserving":[162],"evidence-grounded":[163],"content.":[164],"We":[165],"evaluate":[166],"TLVS":[167],"on":[168],"several":[169],"benchmarks,":[170],"including":[171],"POPE,":[172],"AMBER,":[173],"CHAIR":[174],"(COCO),":[175],"MMHal,":[176],"HallusionBench,":[178],"demonstrating":[179],"consistent":[180],"improvements":[181],"previous":[183],"methods.":[185]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-10T00:00:00"}