{"id":"https://openalex.org/W7134837153","doi":"https://doi.org/10.48550/arxiv.2603.06619","title":"Not all tokens are needed(NAT): token efficient reinforcement learning","display_name":"Not all tokens are needed(NAT): token efficient reinforcement learning","publication_year":2026,"publication_date":"2026-02-20","ids":{"openalex":"https://openalex.org/W7134837153","doi":"https://doi.org/10.48550/arxiv.2603.06619"},"language":null,"primary_location":{"id":"pmh:doi:10.48550/arxiv.2603.06619","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Article"},"type":"article","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":null,"any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5027724386","display_name":"Hejian Sang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sang, Hejian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126942858","display_name":"Yuanda Xu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Yuanda","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128646197","display_name":"Zhengze Zhou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Zhengze","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128671673","display_name":"Ran He","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"He, Ran","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5128661647","display_name":"Zhipeng Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Zhipeng","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":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.26711065,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"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/T10181","display_name":"Natural Language Processing Techniques","score":0.3677999973297119,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.3677999973297119,"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/T10028","display_name":"Topic Modeling","score":0.2849000096321106,"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/T11948","display_name":"Machine Learning in Materials Science","score":0.03370000049471855,"subfield":{"id":"https://openalex.org/subfields/2505","display_name":"Materials Chemistry"},"field":{"id":"https://openalex.org/fields/25","display_name":"Materials 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":0.7592999935150146},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.7045000195503235},{"id":"https://openalex.org/keywords/estimator","display_name":"Estimator","score":0.5128999948501587},{"id":"https://openalex.org/keywords/fraction","display_name":"Fraction (chemistry)","score":0.41819998621940613},{"id":"https://openalex.org/keywords/encoding","display_name":"Encoding (memory)","score":0.41110000014305115},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.3962000012397766},{"id":"https://openalex.org/keywords/core","display_name":"Core (optical fiber)","score":0.3822000026702881},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.35569998621940613},{"id":"https://openalex.org/keywords/sampling","display_name":"Sampling (signal processing)","score":0.3443000018596649}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8001999855041504},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.7592999935150146},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.7045000195503235},{"id":"https://openalex.org/C185429906","wikidata":"https://www.wikidata.org/wiki/Q1130160","display_name":"Estimator","level":2,"score":0.5128999948501587},{"id":"https://openalex.org/C149629883","wikidata":"https://www.wikidata.org/wiki/Q660926","display_name":"Fraction (chemistry)","level":2,"score":0.41819998621940613},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.41749998927116394},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.41110000014305115},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3962000012397766},{"id":"https://openalex.org/C2164484","wikidata":"https://www.wikidata.org/wiki/Q5170150","display_name":"Core (optical fiber)","level":2,"score":0.3822000026702881},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.35569998621940613},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.3443000018596649},{"id":"https://openalex.org/C155032097","wikidata":"https://www.wikidata.org/wiki/Q798503","display_name":"Backpropagation","level":3,"score":0.3361000120639801},{"id":"https://openalex.org/C141603448","wikidata":"https://www.wikidata.org/wiki/Q134830","display_name":"Prefix","level":2,"score":0.3328999876976013},{"id":"https://openalex.org/C55166926","wikidata":"https://www.wikidata.org/wiki/Q2892946","display_name":"Oracle","level":2,"score":0.3287999927997589},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.30329999327659607},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.2994999885559082},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.29910001158714294},{"id":"https://openalex.org/C94966114","wikidata":"https://www.wikidata.org/wiki/Q29256","display_name":"Black box","level":2,"score":0.29600000381469727},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.2808000147342682},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.27059999108314514},{"id":"https://openalex.org/C26263659","wikidata":"https://www.wikidata.org/wiki/Q1886131","display_name":"Random seed","level":3,"score":0.26669999957084656},{"id":"https://openalex.org/C2779843651","wikidata":"https://www.wikidata.org/wiki/Q7390335","display_name":"SIGNAL (programming language)","level":2,"score":0.266400009393692},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.2648000121116638},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2648000121116638},{"id":"https://openalex.org/C73602740","wikidata":"https://www.wikidata.org/wiki/Q7795822","display_name":"Thompson sampling","level":3,"score":0.25459998846054077},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.2531999945640564}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:doi:10.48550/arxiv.2603.06619","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Article"},{"id":"doi:10.48550/arxiv.2603.06619","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.06619","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":"pmh:doi:10.48550/arxiv.2603.06619","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Article"},"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":{"Reinforcement":[0],"learning":[1,91],"(RL)":[2],"has":[3],"become":[4],"a":[5,40,51,64,72,82],"key":[6],"driver":[7],"of":[8,43,85,93,135,167],"progress":[9],"in":[10],"large":[11,41],"language":[12],"models,":[13],"but":[14],"scaling":[15,176],"RL":[16,177,199],"to":[17,175],"long":[18,183],"chain-of-thought":[19],"(CoT)":[20],"trajectories":[21],"is":[22,99],"increasingly":[23],"constrained":[24],"by":[25,182],"backpropagation":[26],"over":[27],"every":[28],"generated":[29,86],"token.":[30],"Even":[31],"with":[32,118],"optimized":[33],"rollout":[34,150],"engines,":[35],"full-token":[36,158],"updates":[37,77],"can":[38],"consume":[39],"fraction":[42],"total":[44],"training":[45,200],"cost,":[46],"turning":[47],"token":[48,70,122],"length":[49],"into":[50],"hidden":[52],"tax":[53],"on":[54],"RL.":[55,95],"We":[56,115],"introduce":[57],"Not":[58],"All":[59],"Tokens":[60],"Are":[61],"Needed":[62],"(NAT),":[63],"unified":[65],"framework":[66],"that":[67],"makes":[68],"the":[69,78,90,146,179],"budget":[71],"first-class":[73],"optimization":[74],"primitive.":[75],"NAT":[76,117,156],"policy":[79],"using":[80,162],"only":[81],"selected":[83],"subset":[84],"tokens":[87],"while":[88,161],"preserving":[89],"signal":[92],"full-sequence":[94],"The":[96],"core":[97],"idea":[98],"an":[100,170],"unbiased":[101],"partial-token":[102],"policy-gradient":[103],"estimator":[104],"via":[105],"Horvitz-Thompson":[106],"reweighting,":[107],"which":[108,136],"ensures":[109],"statistically":[110],"correct":[111],"gradients":[112],"despite":[113],"subsampling.":[114],"instantiate":[116],"two":[119],"simple,":[120],"plug-and-play":[121],"selection":[123],"schemes:":[124],"Uniform":[125],"Random":[126,130],"Sampling":[127],"(URS)":[128],"and":[129,139,142,172,194,197],"Prefix":[131],"Cutting":[132],"(RPC),":[133],"both":[134],"reduce":[137],"forward":[138,196],"backward":[140,198],"compute":[141],"memory":[143,193],"without":[144],"modifying":[145],"reward":[147],"computation":[148],"or":[149],"pipeline.":[151],"Across":[152],"mathematical":[153],"reasoning":[154],"benchmarks,":[155],"matches":[157],"GRPO":[159],"performance":[160],"as":[163,165],"few":[164],"50%":[166],"tokens,":[168],"providing":[169],"efficient":[171],"orthogonal":[173],"pathway":[174],"beyond":[178],"limits":[180],"imposed":[181],"trajectories.":[184],"In":[185],"our":[186],"experiments,":[187],"RPC":[188],"saves":[189],"18%":[190],"peak":[191],"GPU":[192],"29%":[195],"time":[201],"for":[202],"Qwen3-8B.":[203]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-03-11T00:00:00"}