{"id":"https://openalex.org/W7139937237","doi":"https://doi.org/10.48550/arxiv.2603.18533","title":"Balancing the Reasoning Load: Difficulty-Differentiated Policy Optimization with Length Redistribution for Efficient and Robust Reinforcement Learning","display_name":"Balancing the Reasoning Load: Difficulty-Differentiated Policy Optimization with Length Redistribution for Efficient and Robust Reinforcement Learning","publication_year":2026,"publication_date":"2026-03-19","ids":{"openalex":"https://openalex.org/W7139937237","doi":"https://doi.org/10.48550/arxiv.2603.18533"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.18533","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.18533","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.2603.18533","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101376177","display_name":"Yinan Xia","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xia, Yinan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130238281","display_name":"Haotian Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Haotian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5130241468","display_name":"Huiming Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Huiming","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":0.2093999981880188,"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":0.2093999981880188,"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":0.18729999661445618,"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":0.1460999995470047,"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/overconfidence-effect","display_name":"Overconfidence effect","score":0.7350000143051147},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.6814000010490417},{"id":"https://openalex.org/keywords/simple","display_name":"Simple (philosophy)","score":0.526199996471405},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.39579999446868896},{"id":"https://openalex.org/keywords/space","display_name":"Space (punctuation)","score":0.3336000144481659},{"id":"https://openalex.org/keywords/optimization-problem","display_name":"Optimization problem","score":0.3118000030517578}],"concepts":[{"id":"https://openalex.org/C51110983","wikidata":"https://www.wikidata.org/wiki/Q16503490","display_name":"Overconfidence effect","level":2,"score":0.7350000143051147},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6814000010490417},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.675000011920929},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.526199996471405},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.51910001039505},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.39579999446868896},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3587999939918518},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.33559998869895935},{"id":"https://openalex.org/C2778572836","wikidata":"https://www.wikidata.org/wiki/Q380933","display_name":"Space (punctuation)","level":2,"score":0.3336000144481659},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.3118000030517578},{"id":"https://openalex.org/C3018263672","wikidata":"https://www.wikidata.org/wiki/Q1296251","display_name":"Efficient algorithm","level":2,"score":0.288100004196167},{"id":"https://openalex.org/C2987595161","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Optimization algorithm","level":2,"score":0.28760001063346863},{"id":"https://openalex.org/C74080474","wikidata":"https://www.wikidata.org/wiki/Q7305975","display_name":"Redistribution (election)","level":3,"score":0.2815999984741211},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2669000029563904},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.266400009393692}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.18533","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.18533","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.2603.18533","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.18533","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":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Reasoning":[1],"Models":[2],"(LRMs)":[3],"have":[4],"shown":[5],"exceptional":[6],"reasoning":[7],"capabilities,":[8,31],"but":[9,42],"they":[10],"also":[11],"suffer":[12],"from":[13],"the":[14,29,36,75,81,96,105,114,120,136,155,166],"issue":[15],"of":[16,159],"overthinking,":[17],"often":[18],"generating":[19,39],"excessively":[20],"long":[21],"and":[22,69,123,151,157,186],"redundant":[23],"answers.":[24],"For":[25],"problems":[26],"that":[27,66],"exceed":[28],"model's":[30],"LRMs":[32],"tend":[33],"to":[34,48,99,117,162],"exhibit":[35],"overconfidence":[37,76],"phenomenon,":[38],"overly":[40],"short":[41],"incorrect":[43],"answers,":[44],"which":[45,112],"may":[46],"contribute":[47],"suboptimal":[49],"performance.":[50,101],"To":[51],"address":[52],"these":[53,131],"issues,":[54],"we":[55,133],"propose":[56,134],"Difficulty-Differentiated":[57],"Policy":[58],"Optimization":[59],"(DDPO),":[60],"an":[61],"efficient":[62],"reinforcement":[63],"learning":[64],"algorithm":[65],"optimizes":[67],"simple":[68,85],"complex":[70,92],"tasks":[71,86],"separately":[72],"based":[73],"on":[74,130,148],"phenomenon.":[77],"Specifically,":[78],"it":[79,94],"reduces":[80,165],"output":[82],"length":[83,115,122,144,169],"for":[84,91,108,143],"without":[87],"compromising":[88],"accuracy,":[89,111],"while":[90,172],"tasks,":[93],"expands":[95],"exploration":[97],"space":[98],"improve":[100],"We":[102],"further":[103],"derive":[104],"theoretical":[106],"conditions":[107],"maximizing":[109],"expected":[110],"require":[113],"distribution":[116],"closely":[118],"approximate":[119],"optimal":[121],"be":[124],"as":[125,127,139],"concentrated":[126],"possible.":[128],"Based":[129],"conditions,":[132],"using":[135],"difficulty-level":[137],"average":[138,167],"a":[140,181],"well-founded":[141],"reference":[142],"optimization.":[145],"Extensive":[146],"experiments":[147],"both":[149],"in-domain":[150],"out-of-domain":[152],"benchmarks":[153],"validate":[154],"superiority":[156],"effectiveness":[158],"DDPO.":[160],"Compared":[161],"GRPO,":[163],"DDPO":[164],"answer":[168],"by":[170,175],"12%":[171],"improving":[173],"accuracy":[174,185],"1.85%":[176],"across":[177],"multiple":[178],"benchmarks,":[179],"achieving":[180],"better":[182],"trade-off":[183],"between":[184],"length.":[187],"The":[188],"code":[189],"is":[190],"available":[191],"at":[192],"https://github.com/Yinan-Xia/DDPO.":[193]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-03-21T00:00:00"}