{"id":"https://openalex.org/W7202207540","doi":"https://doi.org/10.48550/arxiv.2608.07935","title":"Adaptive Supervised Anchoring for On-Policy Self-Distillation","display_name":"Adaptive Supervised Anchoring for On-Policy Self-Distillation","publication_year":2026,"publication_date":"2026-08-08","ids":{"openalex":"https://openalex.org/W7202207540","doi":"https://doi.org/10.48550/arxiv.2608.07935"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.07935","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.07935","pdf_url":"https://arxiv.org/pdf/2608.07935","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/2608.07935","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5012766373","display_name":"Meilin Yang","orcid":"https://orcid.org/0000-0002-9832-5035","observed_orcids":["https://orcid.org/0000-0002-9832-5035","https://orcid.org/0000-0002-4467-6235"]},"institutions":[{"id":"https://openalex.org/I78988378","display_name":"Renmin University of China","ror":"https://ror.org/041pakw92","country_code":"CN","type":"education","lineage":["https://openalex.org/I78988378"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yang, Meilin","raw_affiliation_strings":["Renmin University of China, Beijing, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Renmin University of China, Beijing, China","institution_ids":["https://openalex.org/I78988378"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101564083","display_name":"Zixuan Ding","orcid":"https://orcid.org/0009-0002-2337-3398","observed_orcids":["https://orcid.org/0009-0002-2337-3398"]},"institutions":[{"id":"https://openalex.org/I78988378","display_name":"Renmin University of China","ror":"https://ror.org/041pakw92","country_code":"CN","type":"education","lineage":["https://openalex.org/I78988378"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Ding, Zixuan","raw_affiliation_strings":["Renmin University of China, Beijing, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Renmin University of China, Beijing, China","institution_ids":["https://openalex.org/I78988378"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126157392","display_name":"Jianhao Nie","orcid":"https://orcid.org/0009-0004-5561-6635","observed_orcids":["https://orcid.org/0009-0004-5561-6635"]},"institutions":[{"id":"https://openalex.org/I78988378","display_name":"Renmin University of China","ror":"https://ror.org/041pakw92","country_code":"CN","type":"education","lineage":["https://openalex.org/I78988378"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Nie, Jianhao","raw_affiliation_strings":["Renmin University of China, Beijing, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Renmin University of China, Beijing, China","institution_ids":["https://openalex.org/I78988378"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5040549480","display_name":"Weite Zhang","orcid":"https://orcid.org/0000-0002-1263-4778","observed_orcids":["https://orcid.org/0000-0002-1263-4778"]},"institutions":[{"id":"https://openalex.org/I78988378","display_name":"Renmin University of China","ror":"https://ror.org/041pakw92","country_code":"CN","type":"education","lineage":["https://openalex.org/I78988378"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhang, Weite","raw_affiliation_strings":["Renmin University of China, Beijing, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Renmin University of China, Beijing, China","institution_ids":["https://openalex.org/I78988378"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136149016","display_name":"Yuxin Zhang","orcid":"https://orcid.org/0009-0004-3083-1101","observed_orcids":["https://orcid.org/0009-0004-3083-1101","https://orcid.org/0009-0000-9475-3254"]},"institutions":[{"id":"https://openalex.org/I78988378","display_name":"Renmin University of China","ror":"https://ror.org/041pakw92","country_code":"CN","type":"education","lineage":["https://openalex.org/I78988378"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhang, Yuxin","raw_affiliation_strings":["Renmin University of China, Beijing, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Renmin University of China, Beijing, China","institution_ids":["https://openalex.org/I78988378"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135929273","display_name":"Zhiming Shao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I78988378","display_name":"Renmin University of China","ror":"https://ror.org/041pakw92","country_code":"CN","type":"education","lineage":["https://openalex.org/I78988378"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shao, Zhiming","raw_affiliation_strings":["Renmin University of China, Beijing, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Renmin University of China, Beijing, China","institution_ids":["https://openalex.org/I78988378"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5146528273","display_name":"Li Yu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I78988378","display_name":"Renmin University of China","ror":"https://ror.org/041pakw92","country_code":"CN","type":"education","lineage":["https://openalex.org/I78988378"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yu, Li","raw_affiliation_strings":["Renmin University of China, Beijing, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Renmin University of China, Beijing, China","institution_ids":["https://openalex.org/I78988378"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5103128811","display_name":"Zhe Fu","orcid":"https://orcid.org/0000-0003-4214-9957","observed_orcids":["https://orcid.org/0000-0003-4214-9957"]},"institutions":[{"id":"https://openalex.org/I78988378","display_name":"Renmin University of China","ror":"https://ror.org/041pakw92","country_code":"CN","type":"education","lineage":["https://openalex.org/I78988378"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Fu, Zhe","raw_affiliation_strings":["Renmin University of China, Beijing, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Renmin University of China, Beijing, China","institution_ids":["https://openalex.org/I78988378"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I78988378"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"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":0.6074000000953674,"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":0.6074000000953674,"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":0.16210000216960907,"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":0.07940000295639038,"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":0.996999979019165},{"id":"https://openalex.org/keywords/teacher-student-knowledge-distillation","display_name":"teacher-student distillation","score":0.6240000128746033},{"id":"https://openalex.org/keywords/language-model-adaptation","display_name":"language model adaptation","score":0.5680000185966492},{"id":"https://openalex.org/keywords/distribution-matching","display_name":"distribution matching","score":0.6930000185966492},{"id":"https://openalex.org/keywords/cold-start","display_name":"cold start","score":0.6919999718666077},{"id":"https://openalex.org/keywords/stability-plasticity-trade-off","display_name":"stability-plasticity trade-off","score":0.9350000023841858},{"id":"https://openalex.org/keywords/general-reasoning","display_name":"general reasoning","score":0.5979999899864197}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7407000064849854},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.6762999892234802},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.6270999908447266},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.6022999882698059},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5752000212669373},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5098000168800354},{"id":"https://openalex.org/C59656382","wikidata":"https://www.wikidata.org/wiki/Q191536","display_name":"Conjunction (astronomy)","level":2,"score":0.4699999988079071},{"id":"https://openalex.org/C18483071","wikidata":"https://www.wikidata.org/wiki/Q168432","display_name":"Anchoring","level":2,"score":0.4194999933242798},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.39100000262260437},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.35190001130104065},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.3424000144004822},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.3402000069618225},{"id":"https://openalex.org/C2779843651","wikidata":"https://www.wikidata.org/wiki/Q7390335","display_name":"SIGNAL (programming language)","level":2,"score":0.3278000056743622},{"id":"https://openalex.org/C52970973","wikidata":"https://www.wikidata.org/wiki/Q2497134","display_name":"Adaptive system","level":2,"score":0.3271999955177307},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.30489999055862427},{"id":"https://openalex.org/C141603448","wikidata":"https://www.wikidata.org/wiki/Q134830","display_name":"Prefix","level":2,"score":0.30250000953674316},{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.2791000008583069},{"id":"https://openalex.org/C174348530","wikidata":"https://www.wikidata.org/wiki/Q188635","display_name":"Bridging (networking)","level":2,"score":0.267300009727478},{"id":"https://openalex.org/C139807058","wikidata":"https://www.wikidata.org/wiki/Q352374","display_name":"Adaptation (eye)","level":2,"score":0.26589998602867126},{"id":"https://openalex.org/C2776291640","wikidata":"https://www.wikidata.org/wiki/Q2912517","display_name":"Value (mathematics)","level":2,"score":0.257099986076355}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.07935","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.07935","pdf_url":"https://arxiv.org/pdf/2608.07935","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.2608.07935","doi":"https://doi.org/10.48550/arxiv.2608.07935","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.07935","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:2608.07935","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.07935","pdf_url":"https://arxiv.org/pdf/2608.07935","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","id":"https://metadata.un.org/sdg/4","score":0.5}],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7202207540.pdf","grobid_xml":"https://content.openalex.org/works/W7202207540.grobid-xml"},"referenced_works_count":12,"referenced_works":["https://openalex.org/W1821462560","https://openalex.org/W2794325560","https://openalex.org/W4380136738","https://openalex.org/W4391631327","https://openalex.org/W4397028161","https://openalex.org/W4414037010","https://openalex.org/W4414939355","https://openalex.org/W4415967437","https://openalex.org/W7080122776","https://openalex.org/W7125861034","https://openalex.org/W7125915072","https://openalex.org/W7161057388"],"related_works":["https://openalex.org/W7204449334","https://openalex.org/W7204908525","https://openalex.org/W7202213474","https://openalex.org/W7204470223","https://openalex.org/W7214887446","https://openalex.org/W7222997000","https://openalex.org/W7172431632","https://openalex.org/W7202177007","https://openalex.org/W7223070783","https://openalex.org/W7160898890"],"abstract_inverted_index":{"On-policy":[0],"self-distillation":[1,184],"(OPSD)":[2],"adapts":[3],"a":[4,11,71,167,179],"language":[5],"model":[6,143],"by":[7],"distilling":[8],"guidance":[9,88,192],"from":[10,17,38,193],"frozen":[12],"teacher":[13,43],"on":[14,25,44,89,100,111],"trajectories":[15],"sampled":[16],"the":[18,26,42,91,105,122,125,153,187],"student.":[19],"Its":[20],"effectiveness,":[21],"however,":[22],"depends":[23],"critically":[24],"quality":[27,138,177],"of":[28,107,124,189],"those":[29],"trajectories.":[30],"We":[31],"show":[32,151],"that":[33,75,152],"when":[34],"student":[35,92],"rollouts":[36],"drift":[37],"target":[39,109],"trajectories,":[40],"conditioning":[41],"off-target":[45],"prefixes":[46],"substantially":[47],"weakens":[48],"its":[49],"task-relevant":[50],"supervision.":[51,195],"Controlled":[52],"prefix-corruption":[53],"experiments":[54],"expose":[55],"this":[56,67],"failure":[57],"mode,":[58],"which":[59],"we":[60,69],"term":[61],"rollout-conditioned":[62,84,191],"signal":[63],"degradation.":[64],"To":[65],"address":[66],"problem,":[68],"propose":[70],"unified":[72],"training":[73],"framework":[74],"separates":[76],"two":[77,145],"complementary":[78],"supervision":[79],"pathways.":[80],"The":[81,95],"first":[82],"retains":[83],"distribution":[85],"matching,":[86],"providing":[87],"states":[90],"actually":[93],"visits.":[94],"second":[96],"applies":[97],"supervised":[98],"cross-entropy":[99],"canonical":[101,194],"ground-truth":[102],"contexts,":[103],"avoiding":[104],"incompatibility":[106],"imposing":[108],"tokens":[110],"erroneous":[112],"rollout":[113,137],"prefixes.":[114],"Token-level":[115],"rollout-target":[116],"alignment":[117],"is":[118],"used":[119],"to":[120],"adapt":[121],"strength":[123],"canonical-context":[126],"anchor,":[127],"emphasizing":[128],"it":[129,135],"during":[130],"cold":[131],"start":[132],"and":[133,148,185],"relaxing":[134],"as":[136,178],"improves.":[139],"Experiments":[140],"across":[141],"multiple":[142],"scales,":[144],"task":[146,157],"families,":[147],"general-reasoning":[149],"benchmarks":[150],"proposed":[154],"approach":[155],"improves":[156],"acquisition":[158],"over":[159],"OPSD":[160],"while":[161],"preserving":[162],"general":[163],"capabilities,":[164],"resulting":[165],"in":[166,182],"more":[168],"favorable":[169],"empirical":[170],"plasticity-stability":[171],"trade-off.":[172],"These":[173],"findings":[174],"identify":[175],"context":[176],"central":[180],"bottleneck":[181],"on-policy":[183],"demonstrate":[186],"value":[188],"separating":[190]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-12T12:28:57.059296","created_date":"2026-08-12T00:00:00"}