{"id":"https://openalex.org/W7204449334","doi":"https://doi.org/10.48550/arxiv.2608.26019","title":"DualOPSD: Closing the Feedback Loop in On-Policy Self-Distillation","display_name":"DualOPSD: Closing the Feedback Loop in On-Policy Self-Distillation","publication_year":2026,"publication_date":"2026-08-26","ids":{"openalex":"https://openalex.org/W7204449334","doi":"https://doi.org/10.48550/arxiv.2608.26019"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.26019","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.26019","pdf_url":"https://arxiv.org/pdf/2608.26019","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/2608.26019","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5004898501","display_name":"Yutong Chen","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I8078737","display_name":"Clemson University","ror":"https://ror.org/037s24f05","country_code":"US","type":"education","lineage":["https://openalex.org/I8078737"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Chen, Yutong","raw_affiliation_strings":["Department of Computer Science , Clemson University 105 Sikes Hall , Clemson , SC 29634 , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , Clemson University 105 Sikes Hall , Clemson , SC 29634 , USA","institution_ids":["https://openalex.org/I8078737"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102911455","display_name":"G.F. Guo","orcid":"https://orcid.org/0009-0005-1270-3736","observed_orcids":["https://orcid.org/0009-0005-1270-3736"]},"institutions":[{"id":"https://openalex.org/I8078737","display_name":"Clemson University","ror":"https://ror.org/037s24f05","country_code":"US","type":"education","lineage":["https://openalex.org/I8078737"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Guo, Guangfu","raw_affiliation_strings":["Department of Computer Science , Clemson University 105 Sikes Hall , Clemson , SC 29634 , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , Clemson University 105 Sikes Hall , Clemson , SC 29634 , USA","institution_ids":["https://openalex.org/I8078737"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5148419427","display_name":"Zhichao Xu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I223532165","display_name":"University of Utah","ror":"https://ror.org/03r0ha626","country_code":"US","type":"education","lineage":["https://openalex.org/I223532165"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Xu, Zhichao","raw_affiliation_strings":["Department of Computer Science , University of Utah"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , University of Utah","institution_ids":["https://openalex.org/I223532165"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5148414271","display_name":"Kunpeng Liu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I8078737","display_name":"Clemson University","ror":"https://ror.org/037s24f05","country_code":"US","type":"education","lineage":["https://openalex.org/I8078737"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Liu, Kunpeng","raw_affiliation_strings":["Department of Computer Science , Clemson University 105 Sikes Hall , Clemson , SC 29634 , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , Clemson University 105 Sikes Hall , Clemson , SC 29634 , USA","institution_ids":["https://openalex.org/I8078737"]}]}],"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":2,"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":0.3521000146865845,"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":0.3521000146865845,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.21789999306201935,"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":0.15479999780654907,"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-self-distillation","display_name":"on-policy self-distillation","score":0.9860000014305115},{"id":"https://openalex.org/keywords/kl-divergence-minimization","display_name":"KL divergence minimization","score":0.5059999823570251},{"id":"https://openalex.org/keywords/teacher-adaptation","display_name":"teacher adaptation","score":0.6779999732971191},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.6129999756813049},{"id":"https://openalex.org/keywords/math-benchmark","display_name":"MATH benchmark","score":0.4399999976158142},{"id":"https://openalex.org/keywords/qwen","display_name":"Qwen","score":0.3840000033378601}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.503000020980835},{"id":"https://openalex.org/C145420912","wikidata":"https://www.wikidata.org/wiki/Q853077","display_name":"Mathematics education","level":1,"score":0.4381999969482422},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.41999998688697815},{"id":"https://openalex.org/C110121322","wikidata":"https://www.wikidata.org/wiki/Q865811","display_name":"Distribution (mathematics)","level":2,"score":0.40130001306533813},{"id":"https://openalex.org/C2776445246","wikidata":"https://www.wikidata.org/wiki/Q1792644","display_name":"Style (visual arts)","level":2,"score":0.39969998598098755},{"id":"https://openalex.org/C19417346","wikidata":"https://www.wikidata.org/wiki/Q7922","display_name":"Pedagogy","level":1,"score":0.3070000112056732},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.289900004863739},{"id":"https://openalex.org/C3020207766","wikidata":"https://www.wikidata.org/wiki/Q4405230","display_name":"Student teacher","level":3,"score":0.28529998660087585},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.27889999747276306},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.2782000005245209},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.26010000705718994}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.26019","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.26019","pdf_url":"https://arxiv.org/pdf/2608.26019","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.2608.26019","doi":"https://doi.org/10.48550/arxiv.2608.26019","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.26019","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:2608.26019","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.26019","pdf_url":"https://arxiv.org/pdf/2608.26019","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":[{"display_name":"Quality education","score":0.8278999924659729,"id":"https://metadata.un.org/sdg/4"}],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7204449334.pdf","grobid_xml":"https://content.openalex.org/works/W7204449334.grobid-xml"},"referenced_works_count":6,"referenced_works":["https://openalex.org/W1821462560","https://openalex.org/W4380993498","https://openalex.org/W4391631327","https://openalex.org/W4414281281","https://openalex.org/W4414939355","https://openalex.org/W7125861034"],"related_works":["https://openalex.org/W7202213474","https://openalex.org/W7202181542","https://openalex.org/W7213073827","https://openalex.org/W7202177007","https://openalex.org/W7204908525","https://openalex.org/W7172431632","https://openalex.org/W7202207540","https://openalex.org/W7214887446","https://openalex.org/W7203880136","https://openalex.org/W7202213136"],"abstract_inverted_index":{"On-policy":[0],"self-distillation":[1],"(OPSD)":[2],"combines":[3],"student":[4,23,40,61,75,78],"generated":[5],"trajectories":[6],"with":[7],"dense":[8],"supervision":[9,38],"from":[10,24,46],"a":[11,56,138],"privileged":[12,69,93,128,143],"teacher":[13,27,32,70,96],"of":[14,92,142],"the":[15,22,30,37,39,47,73,85,90],"same":[16],"model.":[17],"Its":[18],"pointwise":[19],"clipping":[20],"protects":[21],"locally":[25],"incompatible":[26],"directions,":[28],"but":[29],"fixed":[31],"receives":[33],"no":[34],"feedback":[35],"about":[36],"rejects,":[41],"which":[42,52],"limits":[43],"effective":[44,140],"learning":[45],"teacher.":[48],"We":[49],"introduce":[50],"DualOPSD,":[51],"alternates":[53],"updates":[54],"in":[55],"coupled":[57],"teacher-student":[58],"system.":[59],"The":[60,68],"retains":[62],"OPSD's":[63],"clipped":[64],"forward":[65],"KL":[66],"objective.":[67],"adapts":[71],"to":[72,127],"updated":[74],"after":[76],"each":[77],"step.":[79],"A":[80],"residual":[81],"target":[82],"based":[83],"on":[84],"frozen":[86],"initial":[87],"model":[88],"preserves":[89],"effect":[91],"context":[94],"during":[95],"adaptation.":[97],"Across":[98],"Qwen3-8B/4B/1.7B":[99],"and":[100,124],"three":[101],"math":[102],"benchmarks,":[103],"DualOPSD":[104,135],"improves":[105],"avg@12":[106],"accuracy":[107],"over":[108],"OPSD":[109],"by":[110],"9.83":[111],"percentage":[112],"points.":[113],"Further":[114],"diagnostic":[115],"analyses":[116],"show":[117],"lower":[118],"bidirectional":[119],"KL,":[120],"less":[121],"persistent":[122],"mismatch,":[123],"retained":[125],"sensitivity":[126],"reference.":[129],"Together,":[130],"these":[131],"results":[132],"demonstrate":[133],"that":[134],"can":[136],"be":[137],"more":[139],"form":[141],"self-distillation.":[144],"Code":[145],"is":[146],"available":[147],"at:":[148],"https://github.com/on1262/DualOPSD.":[149]},"counts_by_year":[{"year":2026,"cited_by_count":2}],"updated_date":"2026-11-08T12:12:28.789561","created_date":"2026-08-28T00:00:00"}