{"id":"https://openalex.org/W7164017386","doi":"https://doi.org/10.48550/arxiv.2606.07950","title":"The Easy, the Hard, and the Learnable: Confidence and Difficulty-Adaptive Policy Optimization for LLM Reasoning","display_name":"The Easy, the Hard, and the Learnable: Confidence and Difficulty-Adaptive Policy Optimization for LLM Reasoning","publication_year":2026,"publication_date":"2026-06-06","ids":{"openalex":"https://openalex.org/W7164017386","doi":"https://doi.org/10.48550/arxiv.2606.07950"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.07950","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.07950","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.2606.07950","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138246382","display_name":"Zhanke Zhou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Zhanke","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5116869456","display_name":"Xiangyu Lu","orcid":"https://orcid.org/0000-0002-2095-6830"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Xiangyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5079782436","display_name":"Chentao Cao","orcid":"https://orcid.org/0000-0002-3974-3413"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cao, Chentao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5007251379","display_name":"Brando Miranda","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Miranda, Brando","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138207689","display_name":"Tongliang Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Tongliang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138229232","display_name":"Bo Han","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Han, Bo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5138203606","display_name":"Sanmi Koyejo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Koyejo, Sanmi","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/T10028","display_name":"Topic Modeling","score":0.616100013256073,"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":0.616100013256073,"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/T12072","display_name":"Machine Learning and Algorithms","score":0.039900001138448715,"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":0.0340999998152256,"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/security-token","display_name":"Security token","score":0.5274999737739563},{"id":"https://openalex.org/keywords/value","display_name":"Value (mathematics)","score":0.4991999864578247},{"id":"https://openalex.org/keywords/verifiable-secret-sharing","display_name":"Verifiable secret sharing","score":0.42969998717308044},{"id":"https://openalex.org/keywords/bounded-function","display_name":"Bounded function","score":0.4004000127315521},{"id":"https://openalex.org/keywords/sampling","display_name":"Sampling (signal processing)","score":0.37380000948905945},{"id":"https://openalex.org/keywords/tracking","display_name":"Tracking (education)","score":0.3643999993801117},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.33799999952316284}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6686999797821045},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.5274999737739563},{"id":"https://openalex.org/C2776291640","wikidata":"https://www.wikidata.org/wiki/Q2912517","display_name":"Value (mathematics)","level":2,"score":0.4991999864578247},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.461899995803833},{"id":"https://openalex.org/C85847156","wikidata":"https://www.wikidata.org/wiki/Q59015987","display_name":"Verifiable secret sharing","level":3,"score":0.42969998717308044},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.42399999499320984},{"id":"https://openalex.org/C34388435","wikidata":"https://www.wikidata.org/wiki/Q2267362","display_name":"Bounded function","level":2,"score":0.4004000127315521},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.37380000948905945},{"id":"https://openalex.org/C2775936607","wikidata":"https://www.wikidata.org/wiki/Q466845","display_name":"Tracking (education)","level":2,"score":0.3643999993801117},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.33799999952316284},{"id":"https://openalex.org/C44249647","wikidata":"https://www.wikidata.org/wiki/Q208498","display_name":"Confidence interval","level":2,"score":0.31369999051094055},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.30790001153945923},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.30169999599456787},{"id":"https://openalex.org/C148043351","wikidata":"https://www.wikidata.org/wiki/Q4456944","display_name":"Current (fluid)","level":2,"score":0.2944999933242798},{"id":"https://openalex.org/C166052673","wikidata":"https://www.wikidata.org/wiki/Q83021","display_name":"Empirical evidence","level":2,"score":0.29010000824928284},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.28279998898506165},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2797999978065491}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.07950","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.07950","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.2606.07950","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.07950","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":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"RL":[0,145],"with":[1],"verifiable":[2],"rewards":[3],"can":[4],"substantially":[5],"improve":[6],"LLM":[7],"reasoning,":[8],"yet":[9],"standard":[10],"GRPO-style":[11],"training":[12,52],"often":[13],"treats":[14],"easy,":[15],"hard,":[16],"and":[17,24,40,60,79,90,105,117],"learnable":[18,120,129],"questions":[19,121],"alike":[20],"through":[21],"uniform":[22],"sampling":[23],"weighting,":[25],"leading":[26],"to":[27,113],"inefficient":[28],"compute":[29,134],"allocation.":[30],"We":[31],"study":[32],"GRPO":[33],"by":[34,85],"tracking":[35],"token":[36],"log-probabilities,":[37],"group-normalized":[38],"advantages,":[39],"the":[41,68,80,128],"induced":[42],"token-level":[43],"update":[44,72],"weights.":[45],"This":[46],"reveals":[47],"three":[48],"recurring":[49],"dynamics":[50],"as":[51],"proceeds:":[53],"(1)":[54],"confidence":[55,104],"inflation,":[56],"(2)":[57],"advantage":[58],"contraction,":[59],"(3)":[61],"hierarchical":[62],"convergence.":[63],"These":[64],"findings":[65],"suggest":[66],"that":[67],"utility":[69],"of":[70],"each":[71,97],"depends":[73],"strongly":[74],"on":[75],"both":[76],"question":[77,98],"difficulty":[78],"model's":[81],"current":[82],"competence.":[83],"Motivated":[84],"this,":[86],"we":[87],"propose":[88],"Confidence":[89],"Difficulty-adaptive":[91],"Policy":[92],"Optimization":[93],"(CoDaPO),":[94],"which":[95],"assigns":[96],"a":[99,132],"bounded":[100],"value":[101,112],"from":[102],"rollout":[103],"empirical":[106],"difficulty.":[107],"CoDaPO":[108,139],"then":[109],"uses":[110],"this":[111],"reweight":[114],"policy":[115],"updates":[116],"resample":[118],"high-value":[119],"within":[122,127],"mini-batches,":[123],"thereby":[124],"increasing":[125],"discovery":[126],"band":[130],"under":[131],"fixed":[133],"budget.":[135],"Across":[136],"twelve":[137],"benchmarks,":[138],"consistently":[140],"improves":[141],"accuracy":[142],"over":[143],"existing":[144],"methods.":[146],"Our":[147],"code":[148],"is":[149],"publicly":[150],"available":[151],"at":[152],"https://github.com/tmlr-group/CoDaPO.":[153]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-10T00:00:00"}