{"id":"https://openalex.org/W7214718015","doi":"https://doi.org/10.48550/arxiv.2609.34563","title":"Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence","display_name":"Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence","publication_year":2026,"publication_date":"2026-09-28","ids":{"openalex":"https://openalex.org/W7214718015","doi":"https://doi.org/10.48550/arxiv.2609.34563"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.34563","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.34563","pdf_url":"https://arxiv.org/pdf/2609.34563","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.34563","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5153487516","display_name":"Xi Xiao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I32389192","display_name":"University of Alabama at Birmingham","ror":"https://ror.org/008s83205","country_code":"US","type":"education","lineage":["https://openalex.org/I32389192"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Xiao, Xi","raw_affiliation_strings":["Amazon AGI","University of Alabama at Birmingham"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI","institution_ids":[]},{"raw_affiliation_string":"University of Alabama at Birmingham","institution_ids":["https://openalex.org/I32389192"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153470436","display_name":"Tianchen Zhao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Tianchen","raw_affiliation_strings":["Amazon AGI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153536642","display_name":"Youngeun Kim","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kim, Youngeun","raw_affiliation_strings":["Amazon AGI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153467920","display_name":"Zhuowei Li","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Zhuowei","raw_affiliation_strings":["Amazon AGI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100582736","display_name":"Linghan Xu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Linghan","raw_affiliation_strings":["Amazon AGI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153458939","display_name":"Jiaye Wu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Jiaye","raw_affiliation_strings":["Amazon AGI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153409558","display_name":"Zheng Zhang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Zheng","raw_affiliation_strings":["Amazon AGI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153413121","display_name":"Xiang Xu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Xiang","raw_affiliation_strings":["Amazon AGI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153529006","display_name":"Xuanbai Chen","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Xuanbai","raw_affiliation_strings":["Amazon AGI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5000008086","display_name":"Farhan Tejani","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tejani, Farhan","raw_affiliation_strings":["Amazon AGI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129291481","display_name":"Jakub Zablocki","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zablocki, Jakub","raw_affiliation_strings":["Amazon AGI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153460692","display_name":"Julia Xu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Julia","raw_affiliation_strings":["Amazon AGI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5128029793","display_name":"Yifan Xing","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xing, Yifan","raw_affiliation_strings":["Amazon AGI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"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":1.0,"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":1.0,"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/T10028","display_name":"Topic Modeling","score":0.0,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.0,"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/multimodal-large-language-models","display_name":"multimodal large language models","score":0.9240000247955322},{"id":"https://openalex.org/keywords/qwen2-5-vl","display_name":"Qwen2.5-VL","score":0.7149999737739563}],"concepts":[{"id":"https://openalex.org/C70727504","wikidata":"https://www.wikidata.org/wiki/Q1806878","display_name":"Latent class model","level":2,"score":0.641700029373169},{"id":"https://openalex.org/C2777508537","wikidata":"https://www.wikidata.org/wiki/Q7936620","display_name":"Visual reasoning","level":2,"score":0.6187999844551086},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.5486000180244446},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.539900004863739},{"id":"https://openalex.org/C112933361","wikidata":"https://www.wikidata.org/wiki/Q2845258","display_name":"Probabilistic latent semantic analysis","level":2,"score":0.4966000020503998},{"id":"https://openalex.org/C67758918","wikidata":"https://www.wikidata.org/wiki/Q2744408","display_name":"Latent inhibition","level":4,"score":0.4690999984741211},{"id":"https://openalex.org/C100776233","wikidata":"https://www.wikidata.org/wiki/Q2532492","display_name":"Bridge (graph theory)","level":2,"score":0.45680001378059387},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4537999927997589},{"id":"https://openalex.org/C51167844","wikidata":"https://www.wikidata.org/wiki/Q4422623","display_name":"Latent variable","level":2,"score":0.4262000024318695},{"id":"https://openalex.org/C65965080","wikidata":"https://www.wikidata.org/wiki/Q1806885","display_name":"Latent variable model","level":3,"score":0.38119998574256897},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.3707999885082245},{"id":"https://openalex.org/C156216132","wikidata":"https://www.wikidata.org/wiki/Q17099642","display_name":"Local independence","level":4,"score":0.34950000047683716},{"id":"https://openalex.org/C166052673","wikidata":"https://www.wikidata.org/wiki/Q83021","display_name":"Empirical evidence","level":2,"score":0.3125999867916107},{"id":"https://openalex.org/C178253425","wikidata":"https://www.wikidata.org/wiki/Q162668","display_name":"Visual perception","level":3,"score":0.31040000915527344},{"id":"https://openalex.org/C2776502983","wikidata":"https://www.wikidata.org/wiki/Q690182","display_name":"Contrast (vision)","level":2,"score":0.2906000018119812},{"id":"https://openalex.org/C170133592","wikidata":"https://www.wikidata.org/wiki/Q1806883","display_name":"Latent semantic analysis","level":2,"score":0.2849999964237213},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2782999873161316},{"id":"https://openalex.org/C2986089797","wikidata":"https://www.wikidata.org/wiki/Q6501338","display_name":"Visual attention","level":3,"score":0.2768999934196472},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.2743000090122223},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.2718999981880188},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.2680000066757202},{"id":"https://openalex.org/C169900460","wikidata":"https://www.wikidata.org/wiki/Q2200417","display_name":"Cognition","level":2,"score":0.26089999079704285},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.2558000087738037}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.34563","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.34563","pdf_url":"https://arxiv.org/pdf/2609.34563","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.34563","doi":"https://doi.org/10.48550/arxiv.2609.34563","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.34563","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.34563","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.34563","pdf_url":"https://arxiv.org/pdf/2609.34563","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214718015.pdf","grobid_xml":"https://content.openalex.org/works/W7214718015.grobid-xml"},"referenced_works_count":21,"referenced_works":["https://openalex.org/W4407794264","https://openalex.org/W4416053382","https://openalex.org/W4416780398","https://openalex.org/W7127895501","https://openalex.org/W7141460672","https://openalex.org/W7153011030","https://openalex.org/W7155574374","https://openalex.org/W7160867944","https://openalex.org/W7160979862","https://openalex.org/W7161057124","https://openalex.org/W7161697836","https://openalex.org/W7161706067","https://openalex.org/W7163007618","https://openalex.org/W7163159613","https://openalex.org/W7163208102","https://openalex.org/W7163694025","https://openalex.org/W7164911105","https://openalex.org/W7165810306","https://openalex.org/W7172379612","https://openalex.org/W7203672801","https://openalex.org/W7212004172"],"related_works":["https://openalex.org/W4414968495","https://openalex.org/W7161706067","https://openalex.org/W7221102217","https://openalex.org/W7160319041","https://openalex.org/W7161064612","https://openalex.org/W7172379612","https://openalex.org/W7117303718","https://openalex.org/W7163007618","https://openalex.org/W7160867944","https://openalex.org/W7203672801"],"abstract_inverted_index":{"Latent":[0],"visual":[1,104,153,186,218],"reasoning":[2,22,31,187],"(LVR)":[3],"enables":[4],"multimodal":[5],"large":[6],"language":[7],"models":[8],"(MLLMs)":[9],"to":[10,39,68,106,128,142,155,158,184,196,205],"perform":[11],"intermediate":[12],"computation":[13],"in":[14,24,188],"continuous":[15],"latent":[16,30,42,60,63,115,133,189,228],"tokens":[17,43,64],"rather":[18],"than":[19],"expressing":[20],"every":[21],"step":[23],"words.":[25],"However,":[26],"unlike":[27],"textual":[28],"CoT,":[29],"is":[32,147],"not":[33],"directly":[34],"observable,":[35],"making":[36],"it":[37],"difficult":[38],"supervise":[40],"what":[41,103,157],"learn.":[44],"In":[45],"this":[46,79,119],"work,":[47],"we":[48,121,180],"first":[49,183],"conduct":[50],"a":[51,59,95],"thorough":[52],"analysis":[53],"of":[54,85,175],"latent-token":[55,212],"behavior":[56],"and":[57,138,149,151,220],"identify":[58],"evidence-credit":[61],"gap:":[62],"respond":[65],"only":[66],"weakly":[67],"image":[69],"perturbations":[70],"that":[71,78,94,192],"alter":[72],"the":[73,83,129,171,182,225],"correct":[74,137],"answer.":[75],"We":[76],"hypothesize":[77],"issue":[80],"stems":[81],"from":[82],"lack":[84],"explicit":[86],"supervision":[87,127,146],"during":[88],"GRPO":[89],"training.":[90],"These":[91],"findings":[92],"suggest":[93],"final-answer":[96],"reward":[97],"provides":[98],"too":[99],"little":[100],"guidance":[101],"on":[102,177,224],"evidence":[105,154],"preserve":[107],"or":[108],"how":[109],"credit":[110],"should":[111],"be":[112],"assigned":[113],"across":[114],"tokens.":[116,229],"To":[117],"bridge":[118],"gap,":[120],"propose":[122],"ReaLVR,":[123],"which":[124],"brings":[125],"visual-evidence":[126],"model's":[130],"own":[131],"free-running":[132],"trajectories.":[134],"ReaLVR":[135,164],"contrasts":[136],"model-generated":[139],"wrong":[140],"answers":[141],"determine":[143],"where":[144],"stronger":[145,214],"needed,":[148],"relevant":[150,217],"mismatched":[152],"specify":[156],"preserve.":[159],"Across":[160],"three":[161],"model":[162,202],"families,":[163],"consistently":[165],"outperforms":[166],"evaluated":[167],"LVR":[168],"baselines,":[169],"achieving":[170],"highest":[172],"five-task":[173],"average":[174],"63.7%":[176],"Qwen2.5-VL-7B.":[178],"Crucially,":[179],"are":[181],"scale":[185],"space,":[190],"showing":[191],"our":[193],"framework":[194],"continues":[195],"deliver":[197],"robust":[198],"improvements":[199],"at":[200],"frontier":[201],"scales":[203],"up":[204],"235B.":[206],"Further":[207],"analyses":[208],"show":[209],"more":[210],"question-sensitive":[211],"positions,":[213],"alignment":[215],"with":[216],"regions,":[219],"greater":[221],"fixed-context":[222],"dependence":[223],"most":[226],"attended":[227]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-30T00:00:00"}