{"id":"https://openalex.org/W7163590749","doi":"https://doi.org/10.48550/arxiv.2606.04036","title":"Self-Distilled Policy Gradient","display_name":"Self-Distilled Policy Gradient","publication_year":2026,"publication_date":"2026-06-02","ids":{"openalex":"https://openalex.org/W7163590749","doi":"https://doi.org/10.48550/arxiv.2606.04036"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2606.04036","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2606.04036","pdf_url":"https://arxiv.org/pdf/2606.04036","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2606.04036","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137905487","display_name":"Yifeng Liu","orcid":"https://orcid.org/0009-0008-7032-3004","observed_orcids":["https://orcid.org/0009-0008-7032-3004"]},"institutions":[{"id":"https://openalex.org/I161318765","display_name":"University of California, Los Angeles","ror":"https://ror.org/046rm7j60","country_code":"US","type":"education","lineage":["https://openalex.org/I161318765"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Liu, Yifeng","raw_affiliation_strings":["Department of Computer Science , University of California , Los Angeles , CA , USA;"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , University of California , Los Angeles , CA , USA;","institution_ids":["https://openalex.org/I161318765"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101704911","display_name":"Shiyuan Zhang","orcid":"https://orcid.org/0000-0002-5076-029X","observed_orcids":["https://orcid.org/0000-0002-5076-029X","https://orcid.org/0009-0006-8255-0664"]},"institutions":[{"id":"https://openalex.org/I161318765","display_name":"University of California, Los Angeles","ror":"https://ror.org/046rm7j60","country_code":"US","type":"education","lineage":["https://openalex.org/I161318765"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhang, Shiyuan","raw_affiliation_strings":["Department of Computer Science , University of California , Los Angeles , CA , USA;"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , University of California , Los Angeles , CA , USA;","institution_ids":["https://openalex.org/I161318765"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137904917","display_name":"Yifan Zhang (119570)","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I161318765","display_name":"University of California, Los Angeles","ror":"https://ror.org/046rm7j60","country_code":"US","type":"education","lineage":["https://openalex.org/I161318765"]},{"id":"https://openalex.org/I20089843","display_name":"Princeton University","ror":"https://ror.org/00hx57361","country_code":"US","type":"education","lineage":["https://openalex.org/I20089843"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhang, Yifan","raw_affiliation_strings":["Department of Computer Science , University of California , Los Angeles , CA , USA;","Princeton AI Laboratory , Princeton University , Princeton , NJ , USA;"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , University of California , Los Angeles , CA , USA;","institution_ids":["https://openalex.org/I161318765"]},{"raw_affiliation_string":"Princeton AI Laboratory , Princeton University , Princeton , NJ , USA;","institution_ids":["https://openalex.org/I20089843"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5120284571","display_name":"Quanquan Gu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I161318765","display_name":"University of California, Los Angeles","ror":"https://ror.org/046rm7j60","country_code":"US","type":"education","lineage":["https://openalex.org/I161318765"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Gu, Quanquan","raw_affiliation_strings":["Department of Computer Science , University of California , Los Angeles , CA , USA;"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , University of California , Los Angeles , CA , USA;","institution_ids":["https://openalex.org/I161318765"]}]}],"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":0.9980000257492065,"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":0.9980000257492065,"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":0.00019999999494757503,"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/T11574","display_name":"Artificial Intelligence in Games","score":9.999999747378752e-05,"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/self-distillation","display_name":"self-distillation","score":0.8870000243186951},{"id":"https://openalex.org/keywords/policy-gradient","display_name":"policy gradient","score":0.8050000071525574},{"id":"https://openalex.org/keywords/on-policy-learning","display_name":"on-policy learning","score":0.6449999809265137},{"id":"https://openalex.org/keywords/language-models","display_name":"language models","score":0.49900001287460327}],"concepts":[{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6704999804496765},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.640999972820282},{"id":"https://openalex.org/C207390915","wikidata":"https://www.wikidata.org/wiki/Q1230525","display_name":"Divergence (linguistics)","level":2,"score":0.6362000107765198},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.6229000091552734},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.4918999969959259},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.46070000529289246},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3919000029563904},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.33869999647140503},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.31690001487731934},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.28700000047683716},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.28349998593330383},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.27799999713897705},{"id":"https://openalex.org/C101454708","wikidata":"https://www.wikidata.org/wiki/Q17106019","display_name":"Standard Model (mathematical formulation)","level":3,"score":0.26260000467300415}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2606.04036","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2606.04036","pdf_url":"https://arxiv.org/pdf/2606.04036","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2606.04036","doi":"https://doi.org/10.48550/arxiv.2606.04036","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.04036","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:2606.04036","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2606.04036","pdf_url":"https://arxiv.org/pdf/2606.04036","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7163590749.pdf","grobid_xml":"https://content.openalex.org/works/W7163590749.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7148703720","https://openalex.org/W7150828934","https://openalex.org/W7202177007","https://openalex.org/W7161301568","https://openalex.org/W7214931014","https://openalex.org/W7167917130","https://openalex.org/W7172399155","https://openalex.org/W7160513348","https://openalex.org/W7202207540","https://openalex.org/W7223056458"],"abstract_inverted_index":{"On-policy":[0],"self-distillation,":[1,60],"where":[2],"a":[3,16,44],"language":[4],"model":[5],"conditions":[6],"on":[7],"privileged":[8],"context":[9],"to":[10],"supervise":[11],"its":[12],"own":[13],"generations,":[14],"is":[15,80],"promising":[17],"source":[18],"of":[19],"dense":[20],"supervision":[21],"for":[22],"sparse-reward":[23],"reinforcement":[24],"learning.":[25],"Actually,":[26],"it":[27],"can":[28],"be":[29],"instantiated":[30],"as":[31,61,63],"an":[32],"auxiliary":[33],"full-vocabulary":[34,58],"student-to-teacher":[35],"reverse":[36],"Kullback-Leibler":[37],"divergence":[38],"loss.":[39],"We":[40],"therefore":[41],"propose":[42],"SDPG,":[43],"self-distilled":[45],"policy-gradient":[46],"framework":[47],"that":[48],"combines":[49],"group-relative":[50],"verifier":[51],"advantages":[52],"with":[53],"normalized":[54],"standard":[55],"deviation,":[56],"exact":[57],"on-policy":[59],"well":[62],"reference-policy":[64],"KL":[65],"regularization.":[66],"Empirically,":[67],"SDPG":[68],"improves":[69],"stability":[70],"and":[71,75],"performance":[72],"over":[73],"RLVR":[74],"self-distillation":[76],"baselines.":[77],"The":[78],"code":[79],"available":[81],"at":[82],"https://github.com/lauyikfung/SDPG.":[83]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-06-05T00:00:00"}