{"id":"https://openalex.org/W7224087904","doi":"https://doi.org/10.48550/arxiv.2610.36477","title":"Proposing Better Rollouts for On-Policy Distillation","display_name":"Proposing Better Rollouts for On-Policy Distillation","publication_year":2026,"publication_date":"2026-10-27","ids":{"openalex":"https://openalex.org/W7224087904","doi":"https://doi.org/10.48550/arxiv.2610.36477"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.36477","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.36477","pdf_url":"https://arxiv.org/pdf/2610.36477","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/2610.36477","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5159994258","display_name":"Sirui Chen","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I157725225","display_name":"University of Illinois Urbana-Champaign","ror":"https://ror.org/047426m28","country_code":"US","type":"education","lineage":["https://openalex.org/I157725225"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Chen, Sirui","raw_affiliation_strings":["University of Illinois Urbana-Champaign"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Illinois Urbana-Champaign","institution_ids":["https://openalex.org/I157725225"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5160010958","display_name":"Yunzhe Qi","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I157725225","display_name":"University of Illinois Urbana-Champaign","ror":"https://ror.org/047426m28","country_code":"US","type":"education","lineage":["https://openalex.org/I157725225"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Qi, Yunzhe","raw_affiliation_strings":["University of Illinois Urbana-Champaign"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Illinois Urbana-Champaign","institution_ids":["https://openalex.org/I157725225"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5160060966","display_name":"Haobo Xu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I157725225","display_name":"University of Illinois Urbana-Champaign","ror":"https://ror.org/047426m28","country_code":"US","type":"education","lineage":["https://openalex.org/I157725225"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Xu, Haobo","raw_affiliation_strings":["University of Illinois Urbana-Champaign"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Illinois Urbana-Champaign","institution_ids":["https://openalex.org/I157725225"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5051330767","display_name":"Jiaru Zou","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I97018004","display_name":"Stanford University","ror":"https://ror.org/00f54p054","country_code":"US","type":"education","lineage":["https://openalex.org/I97018004"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zou, Jiaru","raw_affiliation_strings":["Stanford University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Stanford University","institution_ids":["https://openalex.org/I97018004"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5160057695","display_name":"Mengting Ai","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I157725225","display_name":"University of Illinois Urbana-Champaign","ror":"https://ror.org/047426m28","country_code":"US","type":"education","lineage":["https://openalex.org/I157725225"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ai, Mengting","raw_affiliation_strings":["University of Illinois Urbana-Champaign"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Illinois Urbana-Champaign","institution_ids":["https://openalex.org/I157725225"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5160009439","display_name":"Zihao Li","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I157725225","display_name":"University of Illinois Urbana-Champaign","ror":"https://ror.org/047426m28","country_code":"US","type":"education","lineage":["https://openalex.org/I157725225"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Li, Zihao","raw_affiliation_strings":["University of Illinois Urbana-Champaign"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Illinois Urbana-Champaign","institution_ids":["https://openalex.org/I157725225"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5159989377","display_name":"Jingrui He","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I157725225","display_name":"University of Illinois Urbana-Champaign","ror":"https://ror.org/047426m28","country_code":"US","type":"education","lineage":["https://openalex.org/I157725225"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"He, Jingrui","raw_affiliation_strings":["University of Illinois Urbana-Champaign"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Illinois Urbana-Champaign","institution_ids":["https://openalex.org/I157725225"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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/T10028","display_name":"Topic Modeling","score":0.4253000020980835,"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/T10028","display_name":"Topic Modeling","score":0.4253000020980835,"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.19140000641345978,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.1412000060081482,"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/on-policy-distillation","display_name":"on-policy distillation","score":0.9990000128746033},{"id":"https://openalex.org/keywords/trust-region-optimization","display_name":"trust-region optimization","score":0.7110000252723694},{"id":"https://openalex.org/keywords/mathematical-reasoning","display_name":"mathematical reasoning","score":0.6660000085830688},{"id":"https://openalex.org/keywords/knowledge-based-reasoning","display_name":"knowledge reasoning","score":0.6990000009536743},{"id":"https://openalex.org/keywords/medical-reasoning","display_name":"medical reasoning","score":0.8830000162124634}],"concepts":[{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.6783999800682068},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6191999912261963},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.5918999910354614},{"id":"https://openalex.org/C207673951","wikidata":"https://www.wikidata.org/wiki/Q1303150","display_name":"Compounding","level":2,"score":0.5475999712944031},{"id":"https://openalex.org/C48677424","wikidata":"https://www.wikidata.org/wiki/Q6888088","display_name":"Mode (computer interface)","level":2,"score":0.46700000762939453},{"id":"https://openalex.org/C21880701","wikidata":"https://www.wikidata.org/wiki/Q2144042","display_name":"Process engineering","level":1,"score":0.38960000872612},{"id":"https://openalex.org/C2779436431","wikidata":"https://www.wikidata.org/wiki/Q30672407","display_name":"Policy learning","level":2,"score":0.27790001034736633},{"id":"https://openalex.org/C2778348673","wikidata":"https://www.wikidata.org/wiki/Q739302","display_name":"Production (economics)","level":2,"score":0.25690001249313354},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.24690000712871552},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.24379999935626984}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.36477","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.36477","pdf_url":"https://arxiv.org/pdf/2610.36477","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.2610.36477","doi":"https://doi.org/10.48550/arxiv.2610.36477","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.36477","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2610.36477","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.36477","pdf_url":"https://arxiv.org/pdf/2610.36477","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":[{"score":0.8278999924659729,"display_name":"Quality education","id":"https://metadata.un.org/sdg/4"}],"study_designs":[],"awards":[{"id":"https://openalex.org/G1453181110","display_name":null,"funder_award_id":"#2138286","funder_id":"https://openalex.org/F4320337377","funder_display_name":"Office of Advanced Cyberinfrastructure"},{"id":"https://openalex.org/G1912064254","display_name":"Track 4:  Advanced CI Coordination Ecosystem: Monitoring and Measurement Services","funder_award_id":"2137603","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G2848635288","display_name":null,"funder_award_id":"2138296","funder_id":"https://openalex.org/F4320332222","funder_display_name":"University of Illinois at Urbana-Champaign"},{"id":"https://openalex.org/G3562386289","display_name":null,"funder_award_id":"#2138296","funder_id":"https://openalex.org/F4320337377","funder_display_name":"Office of Advanced Cyberinfrastructure"},{"id":"https://openalex.org/G4006316252","display_name":"Track 1: ACCESS Resource Allocations Marketplace and Platform Services (RAMPS)","funder_award_id":"2138259","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G4010157413","display_name":null,"funder_award_id":"2138259","funder_id":"https://openalex.org/F4320337377","funder_display_name":"Office of Advanced Cyberinfrastructure"},{"id":"https://openalex.org/G4676224565","display_name":"ACO: An Open CI Ecosystem to Advance Scientific Discovery (OpenCI)","funder_award_id":"2138296","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G509285653","display_name":"Track 3: COre National Ecosystem for CyberinfrasTructure (CONECT)","funder_award_id":"2138307","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G5367127336","display_name":null,"funder_award_id":"#2138307","funder_id":"https://openalex.org/F4320337377","funder_display_name":"Office of Advanced Cyberinfrastructure"},{"id":"https://openalex.org/G5464919783","display_name":null,"funder_award_id":"2137603","funder_id":"https://openalex.org/F4320337377","funder_display_name":"Office of Advanced Cyberinfrastructure"},{"id":"https://openalex.org/G5781048298","display_name":"Category I: Bridging the Gap Between AI/ML Computing Demands and Today's Capabilities","funder_award_id":"2320345","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G6709186444","display_name":null,"funder_award_id":"2137603","funder_id":"https://openalex.org/F4320332222","funder_display_name":"University of Illinois at Urbana-Champaign"},{"id":"https://openalex.org/G6754955425","display_name":null,"funder_award_id":"2138286","funder_id":"https://openalex.org/F4320332222","funder_display_name":"University of Illinois at Urbana-Champaign"},{"id":"https://openalex.org/G7374246537","display_name":"III: Small: RareXplain: A Computational Framework for Explainable Rare Category Analysis","funder_award_id":"2117902","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G8300691027","display_name":"Track 2: Customized Multi-tier Assistance, Training, and Computational Help (MATCH) for End User ACCESS to CI","funder_award_id":"2138286","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G8562358136","display_name":null,"funder_award_id":"2138307","funder_id":"https://openalex.org/F4320332222","funder_display_name":"University of Illinois at Urbana-Champaign"},{"id":"https://openalex.org/G8851268641","display_name":null,"funder_award_id":"2138259","funder_id":"https://openalex.org/F4320332222","funder_display_name":"University of Illinois at Urbana-Champaign"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320332222","display_name":"University of Illinois at Urbana-Champaign","ror":"https://ror.org/047426m28"},{"id":"https://openalex.org/F4320337377","display_name":"Office of Advanced Cyberinfrastructure","ror":"https://ror.org/04nh1dc89"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7224087904.pdf","grobid_xml":"https://content.openalex.org/works/W7224087904.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7220938809","https://openalex.org/W7221072085","https://openalex.org/W7214095543","https://openalex.org/W7161162008","https://openalex.org/W7163029321","https://openalex.org/W7141388625","https://openalex.org/W7163026955","https://openalex.org/W7214975371","https://openalex.org/W7160947271","https://openalex.org/W7161054869"],"abstract_inverted_index":{"On-policy":[0],"distillation":[1,142],"(OPD)":[2],"improves":[3,221],"language-model":[4],"post-training":[5],"by":[6,75,224,235],"letting":[7],"the":[8,20,27,42,59,68,76,100,106,129,141,176],"student":[9,69,93,102,177],"learn":[10],"from":[11,41,158],"its":[12],"own":[13],"sampled":[14],"trajectories":[15,67],"under":[16],"teacher":[17,60,77,82,166,179],"supervision,":[18],"aligning":[19],"training":[21],"process":[22],"with":[23],"inference-time":[24],"behavior.":[25],"However,":[26],"rollout":[28,107,130,156,192,239],"policy":[29,108,131,186,240],"in":[30,103,194],"OPD":[31,126,212,227],"is":[32,171],"usually":[33],"treated":[34],"as":[35,96,133],"fixed:":[36],"each":[37],"round":[38],"samples":[39],"directly":[40],"current":[43],"student.":[44],"We":[45,144],"show":[46,208],"that":[47,124,187,238],"this":[48,189],"choice":[49],"can":[50,84],"create":[51],"a":[52,153,159,172,184,195],"compounding":[53],"teacher--student":[54],"mismatch,":[55],"because":[56],"it":[57],"requires":[58,127],"to":[61,91,246],"provide":[62],"effective":[63,244],"supervision":[64,83],"on":[65,164,200,228],"whichever":[66],"samples,":[70],"including":[71],"ones":[72],"less":[73],"reachable":[74],"itself.":[78],"On":[79,216],"such":[80,95],"trajectories,":[81],"become":[85],"noisy":[86],"or":[87],"even":[88],"misleading,":[89],"leading":[90],"undesirable":[92],"patterns":[94,113],"repetition.":[97],"Moreover,":[98],"since":[99],"updated":[101],"turn":[104],"defines":[105],"of":[109,175],"subsequent":[110],"rounds,":[111],"these":[112],"persist":[114],"and":[115,178,181,204,213],"potentially":[116],"amplify":[117],"over":[118,211,226],"training.":[119],"This":[120],"failure":[121],"mode":[122],"suggests":[123],"improving":[125],"treating":[128],"itself":[132],"an":[134,243],"optimization":[135],"target,":[136],"rather":[137],"than":[138],"only":[139],"modifying":[140],"objective.":[143],"propose":[145],"Mixture":[146],"Proposal":[147],"On-Policy":[148],"Distillation":[149],"(MPOPD),":[150],"which":[151],"derives":[152],"student-anchored,":[154],"teacher-guided":[155],"distribution":[157],"local":[160],"trust-region":[161],"objective":[162],"based":[163],"relative":[165],"support.":[167],"The":[168],"resulting":[169],"target":[170],"geometric":[173],"mixture":[174],"policies,":[180],"we":[182],"introduce":[183],"proposal":[185],"approximates":[188],"mixture,":[190],"enabling":[191],"generation":[193,233],"single":[196],"forward":[197],"pass.":[198],"Experiments":[199],"16":[201],"math,":[202],"knowledge,":[203],"medical":[205],"reasoning":[206],"benchmarks":[207],"consistent":[209],"gains":[210],"recent":[214],"variants.":[215],"math":[217],"reasoning,":[218],"our":[219],"method":[220],"average":[222,232],"accuracy":[223],"6.91":[225],"Qwen3-4B-Base":[229],"while":[230],"reducing":[231],"length":[234],"41%,":[236],"suggesting":[237],"design":[241],"provides":[242],"way":[245],"improve":[247],"OPD.":[248]},"counts_by_year":[],"updated_date":"2026-11-06T12:05:46.655746","created_date":"2026-10-29T00:00:00"}