{"id":"https://openalex.org/W7151334117","doi":"https://doi.org/10.48550/arxiv.2604.04142","title":"OP-GRPO: Efficient Off-Policy GRPO for Flow-Matching Models","display_name":"OP-GRPO: Efficient Off-Policy GRPO for Flow-Matching Models","publication_year":2026,"publication_date":"2026-04-05","ids":{"openalex":"https://openalex.org/W7151334117","doi":"https://doi.org/10.48550/arxiv.2604.04142"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.04142","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.04142","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.2604.04142","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5133120605","display_name":"Liyu Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Zhang, Liyu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133074327","display_name":"Kehan Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Kehan","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5004777855","display_name":"Tingrui Han","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Han, Tingrui","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133144742","display_name":"Tao Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Tao","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5015041571","display_name":"Yuxuan Sheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sheng, Yuxuan","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5068195118","display_name":"Shibo He","orcid":"https://orcid.org/0000-0002-1505-6766"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"He, Shibo","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5133078898","display_name":"Chao Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Chao","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":7,"corresponding_author_ids":["https://openalex.org/A5133120605"],"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":0.685699999332428,"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":0.685699999332428,"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":0.07779999822378159,"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/T10531","display_name":"Advanced Vision and Imaging","score":0.021400000900030136,"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/training","display_name":"Training (meteorology)","score":0.616599977016449},{"id":"https://openalex.org/keywords/sample","display_name":"Sample (material)","score":0.5544999837875366},{"id":"https://openalex.org/keywords/clipping","display_name":"Clipping (morphology)","score":0.5432999730110168},{"id":"https://openalex.org/keywords/reuse","display_name":"Reuse","score":0.4474000036716461},{"id":"https://openalex.org/keywords/quality","display_name":"Quality (philosophy)","score":0.4122999906539917},{"id":"https://openalex.org/keywords/sampling","display_name":"Sampling (signal processing)","score":0.39340001344680786},{"id":"https://openalex.org/keywords/yield","display_name":"Yield (engineering)","score":0.2989000082015991},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.2957000136375427}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7537000179290771},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.616599977016449},{"id":"https://openalex.org/C198531522","wikidata":"https://www.wikidata.org/wiki/Q485146","display_name":"Sample (material)","level":2,"score":0.5544999837875366},{"id":"https://openalex.org/C2776848632","wikidata":"https://www.wikidata.org/wiki/Q853463","display_name":"Clipping (morphology)","level":2,"score":0.5432999730110168},{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.4474000036716461},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.41670000553131104},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.4122999906539917},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.39340001344680786},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3441999852657318},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.30070000886917114},{"id":"https://openalex.org/C134121241","wikidata":"https://www.wikidata.org/wiki/Q899301","display_name":"Yield (engineering)","level":2,"score":0.2989000082015991},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2957000136375427},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.28780001401901245},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.2800000011920929},{"id":"https://openalex.org/C163294075","wikidata":"https://www.wikidata.org/wiki/Q581861","display_name":"Noise reduction","level":2,"score":0.2761000096797943},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.27390000224113464},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.26930001378059387},{"id":"https://openalex.org/C2780440489","wikidata":"https://www.wikidata.org/wiki/Q5227278","display_name":"Data-driven","level":2,"score":0.2612000107765198},{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.25279998779296875},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.25209999084472656},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.25099998712539673}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.04142","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.04142","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.2604.04142","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.04142","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":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Post":[0],"training":[1,28,64,138,146],"via":[2],"GRPO":[3,17,40],"has":[4],"demonstrated":[5],"remarkable":[6],"effectiveness":[7],"in":[8,62,145],"improving":[9],"the":[10,37,69,85,137],"generation":[11,123,150],"quality":[12],"of":[13,87,136],"flow-matching":[14,44],"models.":[15,45],"However,":[16],"suffers":[18],"from":[19],"inherently":[20],"low":[21],"sample":[22],"efficiency":[23,147],"due":[24],"to":[25,67,131],"its":[26],"on-policy":[27],"paradigm.":[29],"To":[30],"address":[31],"this":[32,112],"limitation,":[33],"we":[34,47,76,97],"present":[35],"OP-GRPO,":[36],"first":[38],"Off-Policy":[39],"framework":[41],"tailored":[42],"for":[43,60],"First,":[46],"actively":[48],"select":[49],"high-quality":[50],"trajectories":[51,115],"and":[52,99,110,121],"adaptively":[53],"incorporate":[54],"them":[55],"into":[56],"a":[57,78],"replay":[58],"buffer":[59],"reuse":[61],"subsequent":[63],"iterations.":[65],"Second,":[66],"mitigate":[68,111],"distribution":[70],"shift":[71],"introduced":[72],"by":[73,113],"off-policy":[74,108],"samples,":[75],"propose":[77],"sequence-level":[79],"importance":[80],"sampling":[81],"correction":[82],"that":[83,102],"preserves":[84],"integrity":[86],"GRPO's":[88],"clipping":[89],"mechanism":[90],"while":[91,148],"ensuring":[92],"stable":[93],"policy":[94],"updates.":[95],"Third,":[96],"theoretically":[98],"empirically":[100],"show":[101],"late":[103,117],"denoising":[104],"steps":[105,139],"yield":[106],"ill-conditioned":[107],"ratios,":[109],"truncating":[114],"at":[116],"steps.":[118],"Across":[119],"image":[120],"video":[122],"benchmarks,":[124],"OP-GRPO":[125],"achieves":[126],"comparable":[127],"or":[128],"superior":[129],"performance":[130],"Flow-GRPO":[132],"with":[133],"only":[134],"34.2%":[135],"on":[140],"average,":[141],"yielding":[142],"substantial":[143],"gains":[144],"maintaining":[149],"quality.":[151]},"counts_by_year":[],"updated_date":"2026-04-08T06:07:18.267832","created_date":"2026-04-08T00:00:00"}