{"id":"https://openalex.org/W7155157508","doi":"https://doi.org/10.48550/arxiv.2604.19485","title":"EVPO: Explained Variance Policy Optimization for Adaptive Critic Utilization in LLM Post-Training","display_name":"EVPO: Explained Variance Policy Optimization for Adaptive Critic Utilization in LLM Post-Training","publication_year":2026,"publication_date":"2026-04-21","ids":{"openalex":"https://openalex.org/W7155157508","doi":"https://doi.org/10.48550/arxiv.2604.19485"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.19485","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.19485","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.19485","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5134314849","display_name":"Chengjun Pan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pan, Chengjun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134253515","display_name":"Shichun Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Shichun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5025440428","display_name":"Jiaxu Lin","orcid":"https://orcid.org/0009-0005-1777-1690"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Jiahang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134264262","display_name":"Dingwei Zhu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhu, Dingwei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134292741","display_name":"Jiazheng Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Jiazheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134338860","display_name":"Shihan Dou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dou, Shihan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5050969774","display_name":"Songyang Gao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gao, Songyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134310860","display_name":"Zhenhua Han","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Han, Zhenhua","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134292390","display_name":"Binghai Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Binghai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134216690","display_name":"Rui Zheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zheng, Rui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134353671","display_name":"Xuanjing Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Xuanjing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134297037","display_name":"Tao Gui","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gui, Tao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5123010006","display_name":"Yansong Feng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Feng, Yansong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.8952999711036682,"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":0.8952999711036682,"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/T10653","display_name":"Robot Manipulation and Learning","score":0.012900000438094139,"subfield":{"id":"https://openalex.org/subfields/2207","display_name":"Control and Systems Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12611","display_name":"Neural Networks and Reservoir Computing","score":0.010099999606609344,"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/variance","display_name":"Variance (accounting)","score":0.7003999948501587},{"id":"https://openalex.org/keywords/baseline","display_name":"Baseline (sea)","score":0.6226999759674072},{"id":"https://openalex.org/keywords/kalman-filter","display_name":"Kalman filter","score":0.5968999862670898},{"id":"https://openalex.org/keywords/noise","display_name":"Noise (video)","score":0.5235000252723694},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.4715000092983246},{"id":"https://openalex.org/keywords/simplicity","display_name":"Simplicity","score":0.426800012588501},{"id":"https://openalex.org/keywords/selection","display_name":"Selection (genetic algorithm)","score":0.4154999852180481},{"id":"https://openalex.org/keywords/signal","display_name":"SIGNAL (programming language)","score":0.3691999912261963}],"concepts":[{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.7003999948501587},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.6226999759674072},{"id":"https://openalex.org/C157286648","wikidata":"https://www.wikidata.org/wiki/Q846780","display_name":"Kalman filter","level":2,"score":0.5968999862670898},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.5235000252723694},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5059000253677368},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.4715000092983246},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.45249998569488525},{"id":"https://openalex.org/C2776372474","wikidata":"https://www.wikidata.org/wiki/Q508291","display_name":"Simplicity","level":2,"score":0.426800012588501},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.4154999852180481},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.38359999656677246},{"id":"https://openalex.org/C2779843651","wikidata":"https://www.wikidata.org/wiki/Q7390335","display_name":"SIGNAL (programming language)","level":2,"score":0.3691999912261963},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3343000113964081},{"id":"https://openalex.org/C119043178","wikidata":"https://www.wikidata.org/wiki/Q320723","display_name":"Covariate","level":2,"score":0.3305000066757202},{"id":"https://openalex.org/C2780813799","wikidata":"https://www.wikidata.org/wiki/Q3274237","display_name":"Zero (linguistics)","level":2,"score":0.3271999955177307},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.31679999828338623},{"id":"https://openalex.org/C48103436","wikidata":"https://www.wikidata.org/wiki/Q599031","display_name":"State (computer science)","level":2,"score":0.30469998717308044},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.30250000953674316},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.29670000076293945},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.2867000102996826},{"id":"https://openalex.org/C28761237","wikidata":"https://www.wikidata.org/wiki/Q7805321","display_name":"Time horizon","level":2,"score":0.2833000123500824},{"id":"https://openalex.org/C47446073","wikidata":"https://www.wikidata.org/wiki/Q5165890","display_name":"Control theory (sociology)","level":3,"score":0.2797999978065491},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.27869999408721924},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.27469998598098755},{"id":"https://openalex.org/C8171440","wikidata":"https://www.wikidata.org/wiki/Q903414","display_name":"Steady state (chemistry)","level":2,"score":0.26420000195503235},{"id":"https://openalex.org/C197640229","wikidata":"https://www.wikidata.org/wiki/Q2534066","display_name":"Predictability","level":2,"score":0.2531000077724457}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.19485","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.19485","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.19485","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.19485","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"score":0.43461379408836365,"id":"https://metadata.un.org/sdg/16","display_name":"Peace, Justice and strong institutions"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Reinforcement":[0],"learning":[1],"(RL)":[2],"for":[3,20,31],"LLM":[4],"post-training":[5],"faces":[6],"a":[7,14,18,56,81,105,198],"fundamental":[8],"design":[9],"choice:":[10],"whether":[11],"to":[12,44],"use":[13],"learned":[15,57],"critic":[16,58,117,209],"as":[17,29,80,90],"baseline":[19,78,194],"policy":[21],"optimization.":[22],"Classical":[23],"theory":[24],"favors":[25],"critic-based":[26,153],"methods":[27],"such":[28],"PPO":[30,87,187],"variance":[32,101,162],"reduction,":[33],"yet":[34],"critic-free":[35],"alternatives":[36],"like":[37],"GRPO":[38,89,189],"have":[39],"gained":[40],"widespread":[41],"adoption":[42],"due":[43],"their":[45],"simplicity":[46],"and":[47,88,97,149,154,180,188,213],"competitive":[48],"performance.":[49],"We":[50],"show":[51],"that":[52,63,99,126,204,214],"in":[53],"sparse-reward":[54],"settings,":[55],"can":[59],"inject":[60],"estimation":[61],"noise":[62],"exceeds":[64],"the":[65,94,110,116,164,167,205,215],"state":[66],"signal":[67],"it":[68,127],"captures,":[69],"increasing":[70],"rather":[71],"than":[72,163],"reducing":[73],"advantage":[74,156],"variance.":[75,129],"By":[76],"casting":[77],"selection":[79],"Kalman":[82,95],"filtering":[83],"problem,":[84],"we":[85,134],"unify":[86],"two":[91,168],"extremes":[92],"of":[93,166,191],"gain":[96],"prove":[98],"explained":[100],"(EV),":[102],"computable":[103],"from":[104],"single":[106],"training":[107,147,212],"batch,":[108],"identifies":[109],"exact":[111],"boundary:":[112],"positive":[113],"EV":[114,124,144],"indicates":[115],"reduces":[118],"variance,":[119],"while":[120],"zero":[121,218],"or":[122],"negative":[123],"signals":[125],"inflates":[128],"Building":[130],"on":[131,197],"this":[132],"insight,":[133],"propose":[135],"Explained":[136],"Variance":[137],"Policy":[138],"Optimization":[139],"(EVPO),":[140],"which":[141,192],"monitors":[142],"batch-level":[143],"at":[145,169],"each":[146],"step":[148],"adaptively":[150],"switches":[151],"between":[152],"batch-mean":[155],"estimation,":[157],"provably":[158],"achieving":[159],"no":[160],"greater":[161],"better":[165],"every":[170],"step.":[171],"Across":[172],"four":[173],"tasks":[174],"spanning":[175],"classical":[176],"control,":[177],"agentic":[178],"interaction,":[179],"mathematical":[181],"reasoning,":[182],"EVPO":[183],"consistently":[184],"outperforms":[185],"both":[186],"regardless":[190],"fixed":[193],"is":[195,220],"stronger":[196],"given":[199],"task.":[200],"Further":[201],"analysis":[202],"confirms":[203],"adaptive":[206],"gating":[207],"tracks":[208],"maturation":[210],"over":[211],"theoretically":[216],"derived":[217],"threshold":[219],"empirically":[221],"optimal.":[222]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-04-23T00:00:00"}