{"id":"https://openalex.org/W7133345210","doi":"https://doi.org/10.48550/arxiv.2603.01683","title":"Surgical Post-Training: Proximal On-Policy Distillation for Reasoning with Knowledge Retention","display_name":"Surgical Post-Training: Proximal On-Policy Distillation for Reasoning with Knowledge Retention","publication_year":2026,"publication_date":"2026-03-02","ids":{"openalex":"https://openalex.org/W7133345210","doi":"https://doi.org/10.48550/arxiv.2603.01683"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.01683","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.01683","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.01683","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5128005451","display_name":"Wenye Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Lin, Wenye","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5127908819","display_name":"Kai Han","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Han, Kai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":2,"corresponding_author_ids":["https://openalex.org/A5128005451"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.2168000042438507,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.2168000042438507,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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":0.15860000252723694,"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.156700000166893,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/correctness","display_name":"Correctness","score":0.6965000033378601},{"id":"https://openalex.org/keywords/oracle","display_name":"Oracle","score":0.5479999780654907},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.4618000090122223},{"id":"https://openalex.org/keywords/binary-number","display_name":"Binary number","score":0.43209999799728394},{"id":"https://openalex.org/keywords/regularization","display_name":"Regularization (linguistics)","score":0.41519999504089355},{"id":"https://openalex.org/keywords/rectification","display_name":"Rectification","score":0.3294999897480011}],"concepts":[{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.6965000033378601},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6654000282287598},{"id":"https://openalex.org/C55166926","wikidata":"https://www.wikidata.org/wiki/Q2892946","display_name":"Oracle","level":2,"score":0.5479999780654907},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5256999731063843},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.4618000090122223},{"id":"https://openalex.org/C48372109","wikidata":"https://www.wikidata.org/wiki/Q3913","display_name":"Binary number","level":2,"score":0.43209999799728394},{"id":"https://openalex.org/C2776135515","wikidata":"https://www.wikidata.org/wiki/Q17143721","display_name":"Regularization (linguistics)","level":2,"score":0.41519999504089355},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.33880001306533813},{"id":"https://openalex.org/C50942859","wikidata":"https://www.wikidata.org/wiki/Q4967193","display_name":"Rectification","level":3,"score":0.3294999897480011},{"id":"https://openalex.org/C2779190172","wikidata":"https://www.wikidata.org/wiki/Q4913888","display_name":"Binary data","level":3,"score":0.30160000920295715},{"id":"https://openalex.org/C189430467","wikidata":"https://www.wikidata.org/wiki/Q7293293","display_name":"Ranking (information retrieval)","level":2,"score":0.2962000072002411},{"id":"https://openalex.org/C160920958","wikidata":"https://www.wikidata.org/wiki/Q7662746","display_name":"Synthetic data","level":2,"score":0.2815999984741211},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.27239999175071716},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.26420000195503235},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.25619998574256897}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.01683","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.01683","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.01683","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.01683","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Injecting":[0],"new":[1],"reasoning":[2,68,107],"knowledge":[3,51],"into":[4],"Large":[5],"Language":[6],"Models":[7],"(LLMs)":[8],"via":[9,90],"post-training":[10],"often":[11],"induces":[12],"catastrophic":[13],"forgetting.":[14,30,110],"Recent":[15],"studies":[16],"emphasize":[17],"the":[18,40,152],"importance":[19],"of":[20,76],"on-policy":[21,62,95],"data":[22,79],"but":[23],"suggest":[24],"that":[25,39],"KL-divergence":[26],"fails":[27],"to":[28,66,85],"mitigate":[29],"In":[31],"contrast,":[32],"we":[33],"show,":[34],"both":[35],"analytically":[36],"and":[37,97,108,128],"empirically,":[38],"KL-constrained":[41],"reward":[42],"formulation":[43],"actually":[44],"plays":[45],"a":[46,60,78,99,143],"critical":[47],"role":[48],"in":[49],"retaining":[50],"during":[52],"post-training.":[53],"This":[54],"motivates":[55],"our":[56],"Surgical":[57],"Post-Training":[58],"(SPOT),":[59],"proximal":[61,94],"distillation":[63],"framework":[64],"designed":[65],"optimize":[67],"efficiently":[69],"while":[70],"preserving":[71],"prior":[72],"knowledge.":[73],"SPOT":[74,118,141],"consists":[75],"(1)":[77],"rectification":[80],"pipeline":[81],"employing":[82],"an":[83],"Oracle":[84],"surgically":[86],"correct":[87],"erroneous":[88],"steps":[89],"minimal":[91],"edits,":[92],"generating":[93],"data;":[96],"(2)":[98],"reward-based":[100],"binary":[101],"cross-entropy":[102],"objective":[103],"essential":[104],"for":[105,146],"enhancing":[106],"mitigating":[109],"Empirically,":[111],"with":[112],"only":[113],"4k":[114],"rectified":[115],"math":[116],"pairs,":[117],"improves":[119],"Qwen3-8B's":[120],"accuracy":[121],"by":[122],"6.2%":[123],"on":[124,136],"average":[125],"across":[126],"in-domain":[127],"out-of-domain":[129],"tasks,":[130],"requiring":[131],"merely":[132],"16-minute":[133],"model":[134],"training":[135],"8x":[137],"H800":[138],"GPUs.":[139],"Moreover,":[140],"provides":[142],"superior":[144],"initialization":[145],"subsequent":[147],"reinforcement":[148],"learning,":[149],"significantly":[150],"elevating":[151],"performance":[153],"ceiling.":[154],"Code:":[155],"https://github.com/Visual-AI/SPoT":[156]},"counts_by_year":[],"updated_date":"2026-05-20T08:49:12.498775","created_date":"2026-03-04T00:00:00"}