{"id":"https://openalex.org/W7138943143","doi":"https://doi.org/10.48550/arxiv.2603.15724","title":"Meta-TTRL: A Metacognitive Framework for Self-Improving Test-Time Reinforcement Learning in Unified Multimodal Models","display_name":"Meta-TTRL: A Metacognitive Framework for Self-Improving Test-Time Reinforcement Learning in Unified Multimodal Models","publication_year":2026,"publication_date":"2026-03-16","ids":{"openalex":"https://openalex.org/W7138943143","doi":"https://doi.org/10.48550/arxiv.2603.15724"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.15724","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.15724","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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.15724","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5130126374","display_name":"Lit Sin Tan","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Tan, Lit Sin","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129665287","display_name":"Junzhe Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Junzhe","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123010257","display_name":"Xiaolong Fu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fu, Xiaolong","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5115598343","display_name":"Lichen Ma","orcid":"https://orcid.org/0000-0003-3817-2572"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Lichen","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129876808","display_name":"Junshi Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Junshi","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130165835","display_name":"Jianzhong Shi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shi, Jianzhong","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129779661","display_name":"Yan  (1984 - ) Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Yan","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5130033218","display_name":"Lijie Wen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wen, Lijie","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":8,"corresponding_author_ids":["https://openalex.org/A5130126374"],"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.35850000381469727,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.35850000381469727,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.31520000100135803,"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":0.13019999861717224,"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":0.8270000219345093},{"id":"https://openalex.org/keywords/metacognition","display_name":"Metacognition","score":0.5649999976158142},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.5432999730110168},{"id":"https://openalex.org/keywords/limiting","display_name":"Limiting","score":0.4846999943256378},{"id":"https://openalex.org/keywords/reinforcement","display_name":"Reinforcement","score":0.39879998564720154},{"id":"https://openalex.org/keywords/task-analysis","display_name":"Task analysis","score":0.3644999861717224},{"id":"https://openalex.org/keywords/sampling","display_name":"Sampling (signal processing)","score":0.32899999618530273}],"concepts":[{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.8270000219345093},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6862000226974487},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6051999926567078},{"id":"https://openalex.org/C118147538","wikidata":"https://www.wikidata.org/wiki/Q1126970","display_name":"Metacognition","level":3,"score":0.5649999976158142},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.5498999953269958},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.5432999730110168},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.4846999943256378},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.39879998564720154},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.3644999861717224},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.32899999618530273},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.29829999804496765},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.2874000072479248},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.2851000130176544},{"id":"https://openalex.org/C2777267654","wikidata":"https://www.wikidata.org/wiki/Q3519023","display_name":"Test (biology)","level":2,"score":0.2712000012397766},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.2709999978542328},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.26589998602867126}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.15724","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.15724","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":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.15724","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.15724","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":"article"},"sustainable_development_goals":[{"score":0.5741398930549622,"id":"https://metadata.un.org/sdg/4","display_name":"Quality Education"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Existing":[0],"test-time":[1,49,55,118],"scaling":[2],"(TTS)":[3],"methods":[4],"for":[5,122],"unified":[6],"multimodal":[7],"models":[8],"(UMMs)":[9],"in":[10,125],"text-to-image":[11],"(T2I)":[12],"generation":[13,124],"primarily":[14],"rely":[15],"on":[16,96],"search":[17],"or":[18],"sampling":[19],"strategies":[20],"that":[21,80],"produce":[22],"only":[23],"instance-level":[24],"improvements,":[25],"limiting":[26],"the":[27,65,109,115,144],"ability":[28],"to":[29,113,148],"learn":[30],"from":[31,64],"prior":[32],"inferences":[33],"and":[34,71,91,100],"accumulate":[35],"knowledge":[36],"across":[37,84],"similar":[38],"prompts.":[39],"To":[40],"overcome":[41],"these":[42],"limitations,":[43],"we":[44],"propose":[45],"Meta-TTRL,":[46],"a":[47,131],"metacognitive":[48,137],"reinforcement":[50,119],"learning":[51,120],"framework.":[52],"Meta-TTRL":[53,81],"performs":[54],"parameter":[56],"optimization":[57,146],"guided":[58],"by":[59],"model-intrinsic":[60],"monitoring":[61,140],"signals":[62,141],"derived":[63],"meta-knowledge":[66],"of":[67,117],"UMMs,":[68,87],"achieving":[69,93],"self-improvement":[70],"capability-level":[72],"improvement":[73],"at":[74],"test":[75],"time.":[76],"Extensive":[77],"experiments":[78],"demonstrate":[79],"generalizes":[82],"well":[83],"three":[85],"representative":[86],"including":[88],"Janus-Pro-7B,":[89],"BAGEL,":[90],"Qwen-Image,":[92],"significant":[94],"gains":[95],"compositional":[97],"reasoning":[98],"tasks":[99],"multiple":[101],"T2I":[102,123],"benchmarks":[103],"with":[104,143],"limited":[105],"data.":[106],"We":[107],"provide":[108],"first":[110],"comprehensive":[111],"analysis":[112,128],"investigate":[114],"potential":[116],"(TTRL)":[121],"UMMs.":[126],"Our":[127],"further":[129],"reveals":[130],"key":[132],"insight":[133],"underlying":[134],"effective":[135],"TTRL:":[136],"synergy,":[138],"where":[139],"align":[142],"model's":[145],"regime":[147],"enable":[149],"self-improvement.":[150]},"counts_by_year":[],"updated_date":"2026-03-20T20:54:20.808490","created_date":"2026-03-20T00:00:00"}