{"id":"https://openalex.org/W7148709243","doi":"https://doi.org/10.48550/arxiv.2604.01840","title":"Not All Tokens See Equally: Perception-Grounded Policy Optimization for Large Vision-Language Models","display_name":"Not All Tokens See Equally: Perception-Grounded Policy Optimization for Large Vision-Language Models","publication_year":2026,"publication_date":"2026-04-02","ids":{"openalex":"https://openalex.org/W7148709243","doi":"https://doi.org/10.48550/arxiv.2604.01840"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.01840","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.01840","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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.01840","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5132889361","display_name":"Zekai Ye","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Ye, Zekai","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132911555","display_name":"Qiming Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Qiming","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132879460","display_name":"Xiaocheng Feng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Feng, Xiaocheng","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5039651356","display_name":"Ruihan Chen","orcid":"https://orcid.org/0009-0008-6912-4606"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Ruihan","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132919068","display_name":"Ziming Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Ziming","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5032063939","display_name":"Haoyu Ren","orcid":"https://orcid.org/0000-0002-0241-6507"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ren, Haoyu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132855239","display_name":"Kun Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Kun","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132913308","display_name":"Dandan Tu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tu, Dandan","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5132917844","display_name":"Bing Qin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qin, Bing","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":9,"corresponding_author_ids":["https://openalex.org/A5132889361"],"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":0.9754999876022339,"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":0.9754999876022339,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.006000000052154064,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.002199999988079071,"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/reinforcement-learning","display_name":"Reinforcement learning","score":0.6248000264167786},{"id":"https://openalex.org/keywords/divergence","display_name":"Divergence (linguistics)","score":0.4706999957561493},{"id":"https://openalex.org/keywords/bridge","display_name":"Bridge (graph theory)","score":0.4677000045776367},{"id":"https://openalex.org/keywords/dependency","display_name":"Dependency (UML)","score":0.4440000057220459},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.4377000033855438},{"id":"https://openalex.org/keywords/noise","display_name":"Noise (video)","score":0.4262000024318695},{"id":"https://openalex.org/keywords/encoding","display_name":"Encoding (memory)","score":0.38350000977516174},{"id":"https://openalex.org/keywords/semantics","display_name":"Semantics (computer science)","score":0.36629998683929443}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8001000285148621},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6248000264167786},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6152999997138977},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.47119998931884766},{"id":"https://openalex.org/C207390915","wikidata":"https://www.wikidata.org/wiki/Q1230525","display_name":"Divergence (linguistics)","level":2,"score":0.4706999957561493},{"id":"https://openalex.org/C100776233","wikidata":"https://www.wikidata.org/wiki/Q2532492","display_name":"Bridge (graph theory)","level":2,"score":0.4677000045776367},{"id":"https://openalex.org/C19768560","wikidata":"https://www.wikidata.org/wiki/Q320727","display_name":"Dependency (UML)","level":2,"score":0.4440000057220459},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.4377000033855438},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.4262000024318695},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.38350000977516174},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.36629998683929443},{"id":"https://openalex.org/C2777508537","wikidata":"https://www.wikidata.org/wiki/Q7936620","display_name":"Visual reasoning","level":2,"score":0.3628000020980835},{"id":"https://openalex.org/C85847156","wikidata":"https://www.wikidata.org/wiki/Q59015987","display_name":"Verifiable secret sharing","level":3,"score":0.3596000075340271},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.351500004529953},{"id":"https://openalex.org/C157657479","wikidata":"https://www.wikidata.org/wiki/Q2367247","display_name":"Closed captioning","level":3,"score":0.3334999978542328},{"id":"https://openalex.org/C124304363","wikidata":"https://www.wikidata.org/wiki/Q673661","display_name":"Abstraction","level":2,"score":0.296099990606308},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.2784999907016754},{"id":"https://openalex.org/C2776036281","wikidata":"https://www.wikidata.org/wiki/Q48769818","display_name":"Constraint (computer-aided design)","level":2,"score":0.27379998564720154},{"id":"https://openalex.org/C2778334786","wikidata":"https://www.wikidata.org/wiki/Q1586270","display_name":"Variation (astronomy)","level":2,"score":0.265500009059906},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.25920000672340393},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.25450000166893005},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.250900000333786},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.2502000033855438}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.01840","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.01840","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":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.01840","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.01840","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":"article"},"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":{"While":[0],"Reinforcement":[1],"Learning":[2],"from":[3,18,125],"Verifiable":[4],"Rewards":[5],"(RLVR)":[6],"has":[7],"advanced":[8],"reasoning":[9,139],"in":[10],"Large":[11],"Vision-Language":[12],"Models":[13],"(LVLMs),":[14],"prevailing":[15],"frameworks":[16],"suffer":[17],"a":[19,94,109,168],"foundational":[20],"methodological":[21],"flaw:":[22],"by":[23,146],"distributing":[24],"identical":[25],"advantages":[26,103],"across":[27,135],"all":[28],"generated":[29],"tokens,":[30],"these":[31],"methods":[32],"inherently":[33],"dilute":[34],"the":[35,41,58,66,105,132],"learning":[36,116],"signals":[37,117],"essential":[38],"for":[39,118,171],"optimizing":[40],"critical,":[42],"visually-grounded":[43],"steps":[44],"of":[45,62],"multimodal":[46,138,174],"reasoning.":[47,175],"To":[48],"bridge":[49],"this":[50,78],"gap,":[51],"we":[52,86],"formulate":[53],"\\textit{Token":[54],"Visual":[55],"Dependency},":[56],"quantifying":[57],"causal":[59],"information":[60],"gain":[61],"visual":[63],"inputs":[64],"via":[65],"Kullback-Leibler":[67],"(KL)":[68],"divergence":[69],"between":[70],"visual-conditioned":[71],"and":[72,83,152,165],"text-only":[73],"predictive":[74],"distributions.":[75],"Revealing":[76],"that":[77,100,142,156],"dependency":[79],"is":[80,93],"highly":[81],"sparse":[82],"semantically":[84],"pivotal,":[85],"introduce":[87],"Perception-Grounded":[88],"Policy":[89],"Optimization":[90],"(PGPO),":[91],"which":[92],"novel":[95],"fine-grained":[96],"credit":[97],"assignment":[98],"framework":[99],"dynamically":[101],"reshapes":[102],"at":[104],"token":[106],"level.":[107],"Through":[108],"threshold-gated,":[110],"mass-conserving":[111],"mechanism,":[112],"PGPO":[113,143,157],"actively":[114],"amplifies":[115],"visually-dependent":[119],"tokens":[120],"while":[121],"suppressing":[122],"gradient":[123,160],"noise":[124],"linguistic":[126],"priors.":[127],"Extensive":[128],"experiments":[129],"based":[130],"on":[131,148,180],"Qwen2.5-VL":[133],"series":[134],"seven":[136],"challenging":[137],"benchmarks":[140],"demonstrate":[141],"boosts":[144],"models":[145],"18.7%":[147],"average.":[149],"Both":[150],"theoretical":[151],"empirical":[153],"analyses":[154],"confirm":[155],"effectively":[158],"reduces":[159],"variance,":[161],"prevents":[162],"training":[163],"collapse,":[164],"acts":[166],"as":[167],"potent":[169],"regularizer":[170],"robust,":[172],"perception-grounded":[173],"Code":[176],"will":[177],"be":[178],"released":[179],"https://github.com/Yzk1114/PGPO.":[181]},"counts_by_year":[],"updated_date":"2026-04-10T06:02:16.177343","created_date":"2026-04-04T00:00:00"}