{"id":"https://openalex.org/W7136507606","doi":"https://doi.org/10.48550/arxiv.2603.12893","title":"Finite Difference Flow Optimization for RL Post-Training of Text-to-Image Models","display_name":"Finite Difference Flow Optimization for RL Post-Training of Text-to-Image Models","publication_year":2026,"publication_date":"2026-03-13","ids":{"openalex":"https://openalex.org/W7136507606","doi":"https://doi.org/10.48550/arxiv.2603.12893"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.12893","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.12893","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.2603.12893","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5129557478","display_name":"David McAllister","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"McAllister, David","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5048526734","display_name":"Miika Aittala","orcid":"https://orcid.org/0000-0003-0988-5397"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Aittala, Miika","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5004093189","display_name":"Tero Karras","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Karras, Tero","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5059146062","display_name":"Janne Hellsten","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hellsten, Janne","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5031491881","display_name":"Angjoo Kanazawa","orcid":"https://orcid.org/0000-0003-2592-8430"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kanazawa, Angjoo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5080133323","display_name":"Timo Aila","orcid":"https://orcid.org/0000-0002-9437-4438"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Aila, Timo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5029831111","display_name":"Samuli Laine","orcid":"https://orcid.org/0000-0002-0903-3197"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Laine, Samuli","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":7,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.5672000050544739,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.5672000050544739,"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":0.05310000106692314,"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":0.03610000014305115,"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/reinforcement-learning","display_name":"Reinforcement learning","score":0.7371000051498413},{"id":"https://openalex.org/keywords/sampling","display_name":"Sampling (signal processing)","score":0.6624000072479248},{"id":"https://openalex.org/keywords/variance","display_name":"Variance (accounting)","score":0.5903000235557556},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.5885999798774719},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.5838000178337097},{"id":"https://openalex.org/keywords/quality","display_name":"Quality (philosophy)","score":0.5116999745368958},{"id":"https://openalex.org/keywords/flow","display_name":"Flow (mathematics)","score":0.4952000081539154},{"id":"https://openalex.org/keywords/image","display_name":"Image (mathematics)","score":0.49140000343322754}],"concepts":[{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.7371000051498413},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.6624000072479248},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.592199981212616},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.5903000235557556},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.5885999798774719},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.5838000178337097},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.5116999745368958},{"id":"https://openalex.org/C38349280","wikidata":"https://www.wikidata.org/wiki/Q1434290","display_name":"Flow (mathematics)","level":2,"score":0.4952000081539154},{"id":"https://openalex.org/C115961682","wikidata":"https://www.wikidata.org/wiki/Q860623","display_name":"Image (mathematics)","level":2,"score":0.49140000343322754},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.45829999446868896},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4287000000476837},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.37790000438690186},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.37560001015663147},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.36309999227523804},{"id":"https://openalex.org/C55020928","wikidata":"https://www.wikidata.org/wiki/Q3813865","display_name":"Image quality","level":3,"score":0.3359000086784363},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.31290000677108765},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.3012999892234802},{"id":"https://openalex.org/C162392398","wikidata":"https://www.wikidata.org/wiki/Q272404","display_name":"Finite set","level":2,"score":0.29260000586509705},{"id":"https://openalex.org/C58489278","wikidata":"https://www.wikidata.org/wiki/Q1172284","display_name":"Data set","level":2,"score":0.2793000042438507},{"id":"https://openalex.org/C188116033","wikidata":"https://www.wikidata.org/wiki/Q2664563","display_name":"Q-learning","level":3,"score":0.2775999903678894},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.2671000063419342},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.26330000162124634},{"id":"https://openalex.org/C62644790","wikidata":"https://www.wikidata.org/wiki/Q3454689","display_name":"Variance reduction","level":3,"score":0.2540000081062317}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.12893","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.12893","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.2603.12893","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.12893","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":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Reinforcement":[0],"learning":[1,17],"(RL)":[2],"has":[3],"become":[4],"a":[5,76,87,109],"standard":[6],"technique":[7],"for":[8,102],"post-training":[9],"diffusion-based":[10],"image":[11,28],"synthesis":[12],"models,":[13],"as":[14,27,75,86],"it":[15],"enables":[16],"from":[18],"reward":[19],"signals":[20],"to":[21],"explicitly":[22],"improve":[23],"desirable":[24],"aspects":[25],"such":[26],"quality":[29,100,122],"and":[30,54,98,104,118,123],"prompt":[31,124],"alignment.":[32],"In":[33],"this":[34],"paper,":[35],"we":[36,80],"propose":[37],"an":[38],"online":[39],"RL":[40],"variant":[41],"that":[42,70],"reduces":[43],"the":[44,47,56,60,63,82,106],"variance":[45],"in":[46,59],"model":[48],"updates":[49],"by":[50],"sampling":[51,73,84],"paired":[52],"trajectories":[53],"pulling":[55],"flow":[57],"velocity":[58],"direction":[61],"of":[62,112],"more":[64],"favorable":[65],"image.":[66],"Unlike":[67],"existing":[68],"methods":[69],"treat":[71],"each":[72],"step":[74],"separate":[77],"policy":[78],"action,":[79],"consider":[81],"entire":[83],"process":[85],"single":[88],"action.":[89],"We":[90],"experiment":[91],"with":[92],"both":[93],"high-quality":[94],"vision":[95],"language":[96],"models":[97],"off-the-shelf":[99],"metrics":[101],"rewards,":[103],"evaluate":[105],"outputs":[107],"using":[108],"broad":[110],"set":[111],"metrics.":[113],"Our":[114],"method":[115],"converges":[116],"faster":[117],"yields":[119],"higher":[120],"output":[121],"alignment":[125],"than":[126],"previous":[127],"approaches.":[128]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-03-17T00:00:00"}