{"id":"https://openalex.org/W7165692494","doi":"https://doi.org/10.48550/arxiv.2606.23038","title":"EvoRubrics: Dynamic Rubrics as Rewards via Adversarial Co-Evolution for LLM Reinforcement Learning","display_name":"EvoRubrics: Dynamic Rubrics as Rewards via Adversarial Co-Evolution for LLM Reinforcement Learning","publication_year":2026,"publication_date":"2026-06-22","ids":{"openalex":"https://openalex.org/W7165692494","doi":"https://doi.org/10.48550/arxiv.2606.23038"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.23038","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.23038","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":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.23038","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5009005391","display_name":"Hongxin Ding","orcid":"https://orcid.org/0000-0001-8755-2841"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ding, Hongxin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139181333","display_name":"Baixiang Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Baixiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103512462","display_name":"Y D Fang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fang, Yue","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139157267","display_name":"Weibin Liao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liao, Weibin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139155180","display_name":"Zheng Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Zheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100648800","display_name":"Jinyang Zhang","orcid":"https://orcid.org/0000-0003-0984-3218"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Jinyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139147944","display_name":"Zhijing Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Zhijing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139147868","display_name":"Junfeng Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Junfeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139199972","display_name":"Yasha Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Yasha","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":0.636900007724762,"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":0.636900007724762,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.06419999897480011,"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":0.030700000002980232,"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/rubric","display_name":"Rubric","score":0.9621000289916992},{"id":"https://openalex.org/keywords/discriminative-model","display_name":"Discriminative model","score":0.728600025177002},{"id":"https://openalex.org/keywords/adversarial-system","display_name":"Adversarial system","score":0.7055000066757202},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.6424000263214111},{"id":"https://openalex.org/keywords/generator","display_name":"Generator (circuit theory)","score":0.6330999732017517},{"id":"https://openalex.org/keywords/bridging","display_name":"Bridging (networking)","score":0.34279999136924744}],"concepts":[{"id":"https://openalex.org/C111640148","wikidata":"https://www.wikidata.org/wiki/Q847349","display_name":"Rubric","level":2,"score":0.9621000289916992},{"id":"https://openalex.org/C97931131","wikidata":"https://www.wikidata.org/wiki/Q5282087","display_name":"Discriminative model","level":2,"score":0.728600025177002},{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.7055000066757202},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7027000188827515},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6424000263214111},{"id":"https://openalex.org/C2780992000","wikidata":"https://www.wikidata.org/wiki/Q17016113","display_name":"Generator (circuit theory)","level":3,"score":0.6330999732017517},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.49619999527931213},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4869999885559082},{"id":"https://openalex.org/C174348530","wikidata":"https://www.wikidata.org/wiki/Q188635","display_name":"Bridging (networking)","level":2,"score":0.34279999136924744},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.29409998655319214},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.2939000129699707},{"id":"https://openalex.org/C163258240","wikidata":"https://www.wikidata.org/wiki/Q25342","display_name":"Power (physics)","level":2,"score":0.28929999470710754},{"id":"https://openalex.org/C2781215313","wikidata":"https://www.wikidata.org/wiki/Q3493345","display_name":"SPARK (programming language)","level":2,"score":0.2773999869823456},{"id":"https://openalex.org/C100660578","wikidata":"https://www.wikidata.org/wiki/Q18733","display_name":"Recall","level":2,"score":0.26179999113082886}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.23038","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.23038","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":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.23038","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.23038","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":"Preprint"},"sustainable_development_goals":[{"score":0.727688729763031,"display_name":"Reduced inequalities","id":"https://metadata.un.org/sdg/10"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Rubric-based":[0],"rewards":[1],"offer":[2],"interpretable":[3],"and":[4,48,67,85,116,127,140,175],"fine-grained":[5],"optimization":[6],"signals":[7],"for":[8],"reinforcement":[9],"learning":[10,182],"in":[11,124],"open-ended":[12],"tasks":[13],"where":[14,81],"verifiable":[15],"answers":[16],"are":[17],"unavailable.":[18],"However,":[19],"pre-constructed":[20],"rubrics":[21,69],"remain":[22,114],"static":[23,139],"throughout":[24],"training,":[25],"creating":[26],"a":[27,77,82,86,153,159],"fundamental":[28],"mismatch":[29],"with":[30],"the":[31,41,99,103,107,122],"evolving":[32],"policy:":[33],"fixed":[34],"criteria":[35,112],"gradually":[36],"lose":[37],"discriminative":[38,115],"power":[39],"as":[40,152],"model":[42],"improves,":[43],"leading":[44],"to":[45,113,120],"reward":[46,155],"saturation":[47],"potential":[49],"hacking.":[50],"Recent":[51],"dynamic":[52,141],"rubric":[53,104,108,142],"methods":[54],"partially":[55],"address":[56],"this":[57],"but":[58],"rely":[59],"on":[60],"external":[61,165],"frontier":[62],"models":[63],"or":[64],"ground-truth":[65],"answers,":[66],"update":[68],"only":[70],"at":[71,189],"coarse":[72],"granularity.":[73],"We":[74],"propose":[75],"EvoRubrics,":[76],"co-evolutionary":[78],"RL":[79],"framework":[80],"Policy":[83],"LLM":[84],"Rubric":[87,148],"Generator":[88,149],"jointly":[89],"improve":[90],"through":[91],"adversarial":[92],"interaction":[93],"within":[94],"each":[95],"training":[96],"step.":[97],"As":[98],"policy":[100,123],"improves":[101],"under":[102],"generator's":[105],"guidance,":[106],"generator":[109],"adapts":[110],"its":[111],"informative,":[117],"enabling":[118],"evaluation":[119,176],"track":[121],"real":[125],"time":[126],"naturally":[128],"inducing":[129],"an":[130],"automatic":[131],"curriculum.":[132],"Experiments":[133],"show":[134],"that":[135,171],"EvoRubrics":[136],"consistently":[137],"outperforms":[138],"baselines":[143],"across":[144],"benchmarks.":[145],"The":[146],"learned":[147],"further":[150],"generalizes":[151],"transferable":[154],"model.":[156],"Notably,":[157],"even":[158],"fully":[160],"self-supervised":[161],"variant":[162],"without":[163],"any":[164],"supervision":[166],"achieves":[167],"meaningful":[168],"gains,":[169],"suggesting":[170],"co-evolution":[172],"between":[173],"generation":[174],"alone":[177],"can":[178],"provide":[179],"sufficiently":[180],"rich":[181],"signals.":[183],"Our":[184],"code":[185],"is":[186],"publicly":[187],"available":[188],"https://anonymous.4open.science/r/EvoRubrics-2155/.":[190]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-24T00:00:00"}