{"id":"https://openalex.org/W7163681633","doi":"https://doi.org/10.48550/arxiv.2606.05718","title":"ViCuR: Visual Cues as Recoverable Privilege for Multimodal On-Policy Distillation","display_name":"ViCuR: Visual Cues as Recoverable Privilege for Multimodal On-Policy Distillation","publication_year":2026,"publication_date":"2026-06-04","ids":{"openalex":"https://openalex.org/W7163681633","doi":"https://doi.org/10.48550/arxiv.2606.05718"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.05718","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.05718","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.2606.05718","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5076212343","display_name":"Kanghui Tian","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tian, Kanghui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137989553","display_name":"Siyuan Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Siyuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138004919","display_name":"Ziang Yan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yan, Ziang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137992786","display_name":"Sheng Xia","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xia, Sheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137933100","display_name":"Shuai Dong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dong, Shuai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137930631","display_name":"Yi Wang (32470)","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Yi","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":0.5755000114440918,"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":0.5755000114440918,"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/T11902","display_name":"Intelligent Tutoring Systems and Adaptive Learning","score":0.16539999842643738,"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/T10028","display_name":"Topic Modeling","score":0.0272000003606081,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/privilege","display_name":"Privilege (computing)","score":0.7584999799728394},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.583299994468689},{"id":"https://openalex.org/keywords/imitation","display_name":"Imitation","score":0.5440999865531921},{"id":"https://openalex.org/keywords/sensory-cue","display_name":"Sensory cue","score":0.5202999711036682},{"id":"https://openalex.org/keywords/distillation","display_name":"Distillation","score":0.4507000148296356},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.4271000027656555},{"id":"https://openalex.org/keywords/structured-prediction","display_name":"Structured prediction","score":0.3587000072002411},{"id":"https://openalex.org/keywords/grounded-theory","display_name":"Grounded theory","score":0.3517000079154968}],"concepts":[{"id":"https://openalex.org/C2780138299","wikidata":"https://www.wikidata.org/wiki/Q3404265","display_name":"Privilege (computing)","level":2,"score":0.7584999799728394},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6525999903678894},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.583299994468689},{"id":"https://openalex.org/C126388530","wikidata":"https://www.wikidata.org/wiki/Q1131737","display_name":"Imitation","level":2,"score":0.5440999865531921},{"id":"https://openalex.org/C111370547","wikidata":"https://www.wikidata.org/wiki/Q7451120","display_name":"Sensory cue","level":2,"score":0.5202999711036682},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.47609999775886536},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.4620000123977661},{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.4507000148296356},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.4271000027656555},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.3587000072002411},{"id":"https://openalex.org/C156325361","wikidata":"https://www.wikidata.org/wiki/Q1152864","display_name":"Grounded theory","level":3,"score":0.3517000079154968},{"id":"https://openalex.org/C113843644","wikidata":"https://www.wikidata.org/wiki/Q901882","display_name":"Interface (matter)","level":4,"score":0.3276999890804291},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.31700000166893005},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.31690001487731934},{"id":"https://openalex.org/C4679612","wikidata":"https://www.wikidata.org/wiki/Q866298","display_name":"Aggregate (composite)","level":2,"score":0.31360000371932983},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.29499998688697815},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2867000102996826},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.28630000352859497},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.2750999927520752},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.26930001378059387},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.2583000063896179},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.25760000944137573},{"id":"https://openalex.org/C43091099","wikidata":"https://www.wikidata.org/wiki/Q1067788","display_name":"Through-the-lens metering","level":3,"score":0.25119999051094055}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.05718","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.05718","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.2606.05718","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.05718","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":[{"display_name":"Quality Education","score":0.6409931182861328,"id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"On-policy":[0],"distillation":[1,77],"(OPD)":[2],"improves":[3,157],"reasoning":[4],"by":[5,108,162,180],"training":[6],"a":[7,19,24,30,48,73,116],"student":[8],"on":[9,56,166],"trajectories":[10],"sampled":[11],"from":[12,18,96],"its":[13],"own":[14],"policy":[15],"under":[16],"supervision":[17,53],"teacher.":[20],"In":[21],"multimodal":[22,197],"reasoning,":[23],"common":[25],"extension":[26],"is":[27,106,205],"to":[28,59,128,174],"use":[29],"privileged":[31],"teacher":[32,203,209],"that":[33,79,121],"observes":[34],"training-time-only":[35],"signals":[36,57],"such":[37,44],"as":[38,206,208],"reference":[39],"answers":[40],"or":[41,142],"rationales.":[42],"However,":[43],"answer-side":[45,81],"privilege":[46,82,204],"creates":[47],"train-test":[49],"mismatch:":[50],"the":[51,60,89,97,109,139,189,200],"teacher's":[52],"may":[54],"depend":[55],"unavailable":[58],"student,":[61],"encouraging":[62],"shortcut":[63],"imitation":[64],"rather":[65],"than":[66],"visually":[67,74],"grounded":[68,75],"reasoning.":[69],"We":[70],"propose":[71],"ViCuR,":[72],"privileged-teacher":[76],"framework":[78],"replaces":[80],"with":[83,150,184],"visual":[84,99,131],"cues":[85,93],"(query-related":[86],"evidence":[87,105,132],"in":[88,196],"input).":[90],"Because":[91],"these":[92],"are":[94],"derived":[95],"same":[98],"input":[100],"available":[101],"at":[102,188],"inference,":[103],"their":[104],"recoverable":[107],"student.":[110],"To":[111],"support":[112],"this,":[113],"ViCuR":[114,155],"introduces":[115],"lightweight":[117],"cue":[118],"recovery":[119],"module":[120],"uses":[122],"dedicated":[123],"sink-token":[124],"cross-attention":[125],"during":[126],"prefill":[127],"aggregate":[129],"task-relevant":[130],"into":[133],"an":[134],"internal":[135],"representation,":[136],"without":[137],"changing":[138],"inference":[140],"interface":[141],"requiring":[143],"auxiliary":[144],"cue-generation":[145],"losses.":[146],"Across":[147],"seven":[148],"benchmarks":[149],"Qwen3-VL-2B":[151],"and":[152,164,182],"8B":[153,190],"students,":[154],"consistently":[156],"over":[158],"answer-based":[159],"on-policy":[160,198],"self-distillation":[161],"+1.19":[163],"+1.24":[165],"overall":[167],"average":[168],"performance.":[169],"It":[170],"also":[171],"extends":[172],"naturally":[173],"stronger-teacher":[175],"OPD,":[176],"surpassing":[177],"OPD":[178],"baselines":[179],"+0.64":[181],"+1.08,":[183],"consistent":[185],"out-of-domain":[186],"gains":[187],"scale.":[191],"These":[192],"results":[193],"show":[194],"that,":[195],"distillation,":[199],"design":[201],"of":[202],"important":[207],"strength.":[210]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-06T00:00:00"}