{"id":"https://openalex.org/W7214709239","doi":"https://doi.org/10.48550/arxiv.2609.35505","title":"An RL View of OPD: Least Square Policy Distillation for Sample-Efficient LLM Reasoning","display_name":"An RL View of OPD: Least Square Policy Distillation for Sample-Efficient LLM Reasoning","publication_year":2026,"publication_date":"2026-09-28","ids":{"openalex":"https://openalex.org/W7214709239","doi":"https://doi.org/10.48550/arxiv.2609.35505"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.35505","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.35505","pdf_url":"https://arxiv.org/pdf/2609.35505","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/2609.35505","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5153429193","display_name":"Shangzhe Li","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Li, Shangzhe","raw_affiliation_strings":["University of North Carolina at Chapel Hill ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of North Carolina at Chapel Hill ,","institution_ids":["https://openalex.org/I114027177"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153501462","display_name":"Yuxiao Yang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yang, Yuxiao","raw_affiliation_strings":["University of North Carolina at Chapel Hill ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of North Carolina at Chapel Hill ,","institution_ids":["https://openalex.org/I114027177"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153522901","display_name":"Tianrun Yu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I100005738","display_name":"Brigham Young University","ror":"https://ror.org/047rhhm47","country_code":"US","type":"education","lineage":["https://openalex.org/I100005738"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yu, Tianrun","raw_affiliation_strings":["Brigham Young University , 3"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Brigham Young University , 3","institution_ids":["https://openalex.org/I100005738"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153506478","display_name":"Kaixiang Zhao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I100005738","display_name":"Brigham Young University","ror":"https://ror.org/047rhhm47","country_code":"US","type":"education","lineage":["https://openalex.org/I100005738"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhao, Kaixiang","raw_affiliation_strings":["Brigham Young University , 3"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Brigham Young University , 3","institution_ids":["https://openalex.org/I100005738"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137440951","display_name":"Xiaoyun Wang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Xiaoyun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5007476482","display_name":"Taylor W. Killian","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I100005738","display_name":"Brigham Young University","ror":"https://ror.org/047rhhm47","country_code":"US","type":"education","lineage":["https://openalex.org/I100005738"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Killian, Taylor W.","raw_affiliation_strings":["Brigham Young University , 3"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Brigham Young University , 3","institution_ids":["https://openalex.org/I100005738"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5153483023","display_name":"Weitong Zhang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhang, Weitong","raw_affiliation_strings":["University of North Carolina at Chapel Hill ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of North Carolina at Chapel Hill ,","institution_ids":["https://openalex.org/I114027177"]}]}],"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.8881999850273132,"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.8881999850273132,"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/T10028","display_name":"Topic Modeling","score":0.025800000876188278,"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/T12072","display_name":"Machine Learning and Algorithms","score":0.022299999371170998,"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":0.9890000224113464},{"id":"https://openalex.org/keywords/policy-diversity","display_name":"policy diversity","score":0.5950000286102295},{"id":"https://openalex.org/keywords/regret-bounds","display_name":"regret bounds","score":0.7580000162124634},{"id":"https://openalex.org/keywords/mathematical-reasoning","display_name":"mathematical reasoning","score":0.7480000257492065},{"id":"https://openalex.org/keywords/sample-efficiency","display_name":"sample efficiency","score":0.5820000171661377},{"id":"https://openalex.org/keywords/value-based-reinforcement-learning","display_name":"value-based reinforcement learning","score":0.5960000157356262}],"concepts":[{"id":"https://openalex.org/C50817715","wikidata":"https://www.wikidata.org/wiki/Q79895177","display_name":"Regret","level":2,"score":0.7767000198364258},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6780999898910522},{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.657800018787384},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.6424999833106995},{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.6097000241279602},{"id":"https://openalex.org/C527412718","wikidata":"https://www.wikidata.org/wiki/Q855395","display_name":"Interpretation (philosophy)","level":2,"score":0.5511000156402588},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.5458999872207642},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.38510000705718994},{"id":"https://openalex.org/C2781316041","wikidata":"https://www.wikidata.org/wiki/Q1230584","display_name":"Diversity (politics)","level":2,"score":0.3650999963283539},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3264999985694885},{"id":"https://openalex.org/C2779436431","wikidata":"https://www.wikidata.org/wiki/Q30672407","display_name":"Policy learning","level":2,"score":0.32280001044273376},{"id":"https://openalex.org/C13355873","wikidata":"https://www.wikidata.org/wiki/Q2920850","display_name":"Connection (principal bundle)","level":2,"score":0.32100000977516174},{"id":"https://openalex.org/C2780551164","wikidata":"https://www.wikidata.org/wiki/Q2306599","display_name":"Column (typography)","level":3,"score":0.30790001153945923},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.2809999883174896},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2800999879837036},{"id":"https://openalex.org/C149629883","wikidata":"https://www.wikidata.org/wiki/Q660926","display_name":"Fraction (chemistry)","level":2,"score":0.27379998564720154},{"id":"https://openalex.org/C154030694","wikidata":"https://www.wikidata.org/wiki/Q1436074","display_name":"Fractionating column","level":3,"score":0.2621999979019165},{"id":"https://openalex.org/C48677424","wikidata":"https://www.wikidata.org/wiki/Q6888088","display_name":"Mode (computer interface)","level":2,"score":0.2587999999523163},{"id":"https://openalex.org/C206345919","wikidata":"https://www.wikidata.org/wiki/Q20380951","display_name":"Resource (disambiguation)","level":2,"score":0.25529998540878296},{"id":"https://openalex.org/C123587114","wikidata":"https://www.wikidata.org/wiki/Q2101508","display_name":"Policy analysis","level":2,"score":0.2540000081062317}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.35505","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.35505","pdf_url":"https://arxiv.org/pdf/2609.35505","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.2609.35505","doi":"https://doi.org/10.48550/arxiv.2609.35505","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.35505","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.35505","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.35505","pdf_url":"https://arxiv.org/pdf/2609.35505","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":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214709239.pdf","grobid_xml":"https://content.openalex.org/works/W7214709239.grobid-xml"},"referenced_works_count":20,"referenced_works":["https://openalex.org/W1821462560","https://openalex.org/W2609650878","https://openalex.org/W2736601468","https://openalex.org/W2904246096","https://openalex.org/W3083410900","https://openalex.org/W3134642945","https://openalex.org/W3177813494","https://openalex.org/W4388926373","https://openalex.org/W4391631327","https://openalex.org/W4414281281","https://openalex.org/W4414939355","https://openalex.org/W4415566463","https://openalex.org/W7128819059","https://openalex.org/W7134813685","https://openalex.org/W7154505894","https://openalex.org/W7163131677","https://openalex.org/W7164915140","https://openalex.org/W7166885264","https://openalex.org/W7208838824","https://openalex.org/W7213609244"],"related_works":["https://openalex.org/W7171930018","https://openalex.org/W7161165287","https://openalex.org/W7125861034","https://openalex.org/W7203935495","https://openalex.org/W7168289780","https://openalex.org/W7169504124","https://openalex.org/W7168964786","https://openalex.org/W7163891415","https://openalex.org/W7161054869","https://openalex.org/W7220404346"],"abstract_inverted_index":{"We":[0],"study":[1],"on-policy":[2],"distillation":[3,99],"(OPD)":[4],"through":[5,55,119],"the":[6,15,152],"lens":[7],"of":[8,113,155],"reinforcement":[9],"learning,":[10],"establishing":[11],"a":[12,84,170,174],"connection":[13],"between":[14],"reverse-KL":[16],"objective":[17],"in":[18,116],"OPD":[19,149,166],"and":[20,41,77,106,173,180],"KL-regularized":[21],"policy":[22,49,53,131],"optimization.":[23],"Building":[24],"on":[25,165],"this":[26],"connection,":[27],"we":[28,125],"introduce":[29],"Least-Square":[30],"Policy":[31],"Distillation":[32],"(LSPD),":[33],"an":[34,162],"RL-inspired":[35],"framework":[36],"that":[37,79,127,167],"brings":[38],"optimistic":[39,74],"exploration":[40,56],"off-policy":[42,142],"data":[43],"reuse":[44],"from":[45,64],"value-based":[46,75],"RL":[47,163],"into":[48],"distillation.":[50,184],"LSPD":[51,72,95,128],"preserves":[52,130],"diversity":[54,132],"while":[57],"improving":[58],"rollout":[59,156],"efficiency":[60],"by":[61,133],"repeatedly":[62],"learning":[63,76],"previously":[65],"collected":[66],"trajectories.":[67],"Our":[68],"theoretical":[69],"analysis":[70],"connects":[71],"to":[73,123,147],"shows":[78],"its":[80],"idealized":[81],"formulation":[82],"achieves":[83,144],"sharp":[85],"$\\tilde{\\mathcal":[86],"O}(\\log":[87],"K)$":[88],"regret":[89],"bound":[90],"under":[91],"online":[92],"exploration.":[93],"Empirically,":[94],"consistently":[96],"outperforms":[97],"existing":[98],"baselines":[100],"across":[101],"six":[102],"mathematical":[103],"reasoning":[104],"benchmarks":[105],"diverse":[107],"teacher-student":[108],"settings,":[109],"with":[110],"average":[111],"gains":[112],"+1.59":[114],"points":[115],"Avg@16.":[117],"Remarkably,":[118],"Pass@k":[120],"evaluations":[121],"up":[122],"k=64,":[124],"found":[126],"better":[129],"achieving":[134],"stronger":[135],"performance":[136,146],"as":[137],"k":[138],"grows.":[139],"Its":[140],"fully":[141],"variant":[143],"comparable":[145],"vanilla":[148],"using":[150],"only":[151],"first":[153],"25%":[154],"batches.":[157],"Together,":[158],"these":[159],"results":[160],"provide":[161],"perspective":[164],"offers":[168],"both":[169],"principled":[171],"interpretation":[172],"practical":[175],"route":[176],"toward":[177],"more":[178],"effective":[179],"rollout-efficient":[181],"language":[182],"model":[183]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-30T00:00:00"}