{"id":"https://openalex.org/W7160876190","doi":"https://doi.org/10.48550/arxiv.2605.07579","title":"Your Language Model is Its Own Critic: Reinforcement Learning with Value Estimation from Actor's Internal States","display_name":"Your Language Model is Its Own Critic: Reinforcement Learning with Value Estimation from Actor's Internal States","publication_year":2026,"publication_date":"2026-05-08","ids":{"openalex":"https://openalex.org/W7160876190","doi":"https://doi.org/10.48550/arxiv.2605.07579"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.07579","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.07579","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,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.07579","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5135828528","display_name":"Yunho Choi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Choi, Yunho","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135879736","display_name":"Jongwon Lim","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lim, Jongwon","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135847009","display_name":"Woojin Ahn","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ahn, Woojin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135865110","display_name":"Minjae Oh","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Oh, Minjae","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5033034330","display_name":"Jeonghoon Shim","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shim, Jeonghoon","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135827884","display_name":"Yohan Jo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jo, Yohan","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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.3612000048160553,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.3612000048160553,"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":0.28619998693466187,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.0737999975681305,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/estimator","display_name":"Estimator","score":0.7389000058174133},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.7339000105857849},{"id":"https://openalex.org/keywords/verifiable-secret-sharing","display_name":"Verifiable secret sharing","score":0.6305999755859375},{"id":"https://openalex.org/keywords/variance","display_name":"Variance (accounting)","score":0.6053000092506409},{"id":"https://openalex.org/keywords/value","display_name":"Value (mathematics)","score":0.5328999757766724},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.4537999927997589},{"id":"https://openalex.org/keywords/sampling","display_name":"Sampling (signal processing)","score":0.4189000129699707}],"concepts":[{"id":"https://openalex.org/C185429906","wikidata":"https://www.wikidata.org/wiki/Q1130160","display_name":"Estimator","level":2,"score":0.7389000058174133},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.7339000105857849},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.676800012588501},{"id":"https://openalex.org/C85847156","wikidata":"https://www.wikidata.org/wiki/Q59015987","display_name":"Verifiable secret sharing","level":3,"score":0.6305999755859375},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.6053000092506409},{"id":"https://openalex.org/C2776291640","wikidata":"https://www.wikidata.org/wiki/Q2912517","display_name":"Value (mathematics)","level":2,"score":0.5328999757766724},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.4537999927997589},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.4189000129699707},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3824999928474426},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.3785000145435333},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.361299991607666},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.34940001368522644},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.34439998865127563},{"id":"https://openalex.org/C201236551","wikidata":"https://www.wikidata.org/wiki/Q7310809","display_name":"Relative value","level":2,"score":0.31119999289512634},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.287200003862381},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.27639999985694885},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2676999866962433},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2621999979019165},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.2524000108242035},{"id":"https://openalex.org/C62644790","wikidata":"https://www.wikidata.org/wiki/Q3454689","display_name":"Variance reduction","level":3,"score":0.25130000710487366}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.07579","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.07579","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,"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":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.07579","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.07579","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,"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"},"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":{"Reinforcement":[0],"learning":[1,157],"with":[2,48],"verifiable":[3,80,203],"rewards":[4],"(RLVR)":[5],"for":[6,14,143,154,167],"Large":[7],"Reasoning":[8],"Models":[9],"hinges":[10],"on":[11],"baseline":[12,56],"estimation":[13],"variance":[15,153],"reduction,":[16],"but":[17],"existing":[18],"approaches":[19],"pay":[20],"a":[21,26,55,114,135,144,194],"heavy":[22],"price:":[23],"PPO":[24],"requires":[25],"policy-model":[27],"scale":[28],"critic,":[29],"while":[30,182],"GRPO":[31],"needs":[32],"multiple":[33],"rollouts":[34],"per":[35],"prompt":[36,88,131,141],"to":[37,193,201],"keep":[38],"its":[39,187],"empirical":[40],"group":[41],"mean":[42],"stable.":[43],"We":[44],"introduce":[45,113],"Policy":[46],"Optimization":[47],"Internal":[49],"State":[50],"Value":[51],"Estimation),":[52],"which":[53],"obtains":[54],"at":[57],"negligible":[58],"cost":[59],"by":[60],"using":[61,109,133],"the":[62,70,78,83,87,102,161,207],"policy":[63,71,218],"model's":[64,208],"internal":[65,126,210],"signals":[66],"already":[67],"computed":[68],"during":[69,148],"forward":[72],"pass.":[73],"A":[74],"lightweight":[75],"probe":[76],"predicts":[77,118],"expected":[79],"reward":[81],"from":[82,122],"hidden":[84],"states":[85],"of":[86,164],"and":[89,97,158,173,199,216],"generated":[90],"trajectory,":[91],"as":[92,94],"well":[93],"token-entropy":[95],"statistics,":[96],"is":[98],"trained":[99],"online":[100],"alongside":[101],"policy.":[103],"To":[104],"preserve":[105],"gradient":[106,152],"unbiasedness":[107],"despite":[108],"trajectory-conditioned":[110],"features,":[111],"we":[112],"cross-rollout":[115],"construction":[116],"that":[117],"each":[119],"rollout's":[120,125],"value":[121,132,188,197],"an":[123],"independent":[124],"states.":[127],"Because":[128],"POISE":[129,179,212],"estimates":[130],"only":[134],"single":[136],"rollout,":[137],"it":[138],"enables":[139,213],"higher":[140],"diversity":[142],"fixed":[145],"compute":[146,162],"budget":[147],"training.":[149],"This":[150],"reduces":[151],"more":[155,214],"stable":[156,215],"also":[159],"eliminates":[160],"overhead":[163],"sampling":[165],"costs":[166],"detecting":[168],"zero-advantage":[169],"prompts.":[170],"On":[171],"Qwen3-4B":[172],"DeepSeek-R1-Distill-Qwen-1.5B":[174],"across":[175],"math":[176],"reasoning":[177],"benchmarks,":[178],"matches":[180],"DAPO":[181],"requiring":[183],"less":[184],"compute.":[185],"Moreover,":[186],"estimator":[189],"shows":[190],"similar":[191],"performance":[192],"separate":[195],"LLM-scale":[196],"model":[198],"generalizes":[200],"various":[202],"tasks.":[204],"By":[205],"leveraging":[206],"own":[209],"representations,":[211],"efficient":[217],"optimization.":[219]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-12T00:00:00"}