{"id":"https://openalex.org/W7128742332","doi":"https://doi.org/10.48550/arxiv.2602.10520","title":"Prioritize the Process, Not Just the Outcome: Rewarding Latent Thought Trajectories Improves Reasoning in Looped Language Models","display_name":"Prioritize the Process, Not Just the Outcome: Rewarding Latent Thought Trajectories Improves Reasoning in Looped Language Models","publication_year":2026,"publication_date":"2026-02-11","ids":{"openalex":"https://openalex.org/W7128742332","doi":"https://doi.org/10.48550/arxiv.2602.10520"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2602.10520","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2602.10520","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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2602.10520","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5125716952","display_name":"Williams Jonathan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Williams, Jonathan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5125748530","display_name":"Tureci Esin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tureci, Esin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":2,"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":0.33000001311302185,"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":0.33000001311302185,"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":0.13660000264644623,"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/T13702","display_name":"Machine Learning in Healthcare","score":0.07039999961853027,"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/inference","display_name":"Inference","score":0.7091000080108643},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.6516000032424927},{"id":"https://openalex.org/keywords/reinforcement","display_name":"Reinforcement","score":0.39500001072883606},{"id":"https://openalex.org/keywords/commonsense-reasoning","display_name":"Commonsense reasoning","score":0.3659999966621399},{"id":"https://openalex.org/keywords/latent-variable","display_name":"Latent variable","score":0.3456000089645386},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.34310001134872437}],"concepts":[{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.7091000080108643},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6516000032424927},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6205999851226807},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5867000222206116},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4921000003814697},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.39500001072883606},{"id":"https://openalex.org/C193221554","wikidata":"https://www.wikidata.org/wiki/Q5153664","display_name":"Commonsense reasoning","level":2,"score":0.3659999966621399},{"id":"https://openalex.org/C51167844","wikidata":"https://www.wikidata.org/wiki/Q4422623","display_name":"Latent variable","level":2,"score":0.3456000089645386},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.34310001134872437},{"id":"https://openalex.org/C83725634","wikidata":"https://www.wikidata.org/wiki/Q7268699","display_name":"Qualitative reasoning","level":2,"score":0.32519999146461487},{"id":"https://openalex.org/C28427503","wikidata":"https://www.wikidata.org/wiki/Q13580300","display_name":"Internal model","level":3,"score":0.3046000003814697},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.29120001196861267},{"id":"https://openalex.org/C2781384469","wikidata":"https://www.wikidata.org/wiki/Q6786277","display_name":"Matching law","level":3,"score":0.27790001034736633},{"id":"https://openalex.org/C65965080","wikidata":"https://www.wikidata.org/wiki/Q1806885","display_name":"Latent variable model","level":3,"score":0.2694000005722046}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2602.10520","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2602.10520","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":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2602.10520","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2602.10520","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":"article"},"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":{"Looped":[0],"Language":[1],"Models":[2],"(LoopLMs)":[3],"perform":[4],"multi-step":[5],"latent":[6,51,82],"reasoning":[7,17,29,83,125,159],"prior":[8],"to":[9,25,48,157],"token":[10],"generation":[11],"and":[12,96,112,133,141],"outperform":[13],"conventional":[14],"LLMs":[15],"on":[16,93,122,137,143,151],"benchmarks":[18],"at":[19,176],"smaller":[20],"parameter":[21],"budgets.":[22],"However,":[23],"attempts":[24],"further":[26],"improve":[27],"LoopLM":[28],"with":[30,57,101,107],"reinforcement":[31,73,169],"learning":[32,74,170],"have":[33],"failed":[34],"-":[35],"standard":[36],"objectives":[37],"such":[38],"as":[39],"Group":[40],"Relative":[41],"Policy":[42],"Optimization":[43],"(GRPO)":[44],"only":[45],"assign":[46],"credit":[47,89,166],"the":[49,58,80,138,144,162],"final":[50],"state,":[52],"creating":[53],"a":[54,72],"fundamental":[55],"mismatch":[56],"model's":[59],"internal":[60],"computation.":[61],"To":[62],"resolve":[63],"this,":[64],"we":[65],"introduce":[66],"RLTT":[67,85,115,153],"(Reward":[68],"Latent":[69],"Thought":[70],"Trajectories),":[71],"framework":[75],"which":[76],"distributes":[77],"reward":[78],"across":[79],"full":[81],"trajectory.":[84],"provides":[86],"dense,":[87],"trajectory-level":[88,165],"assignment":[90,167],"without":[91],"relying":[92],"external":[94],"verifiers":[95],"can":[97],"directly":[98],"replace":[99],"GRPO":[100,121],"negligible":[102],"overhead.":[103],"Across":[104],"extensive":[105],"experiments":[106],"Ouro-1.4B/2.6B-Thinking":[108],"under":[109],"identical":[110],"training":[111],"inference":[113],"conditions,":[114],"yields":[116],"statistically":[117],"significant":[118],"improvements":[119],"over":[120,130],"challenging":[123],"mathematical":[124],"benchmarks,":[126,160],"improving":[127],"mean":[128],"accuracy":[129],"MATH-500,":[131],"AIME24/26,":[132],"BeyondAIME":[134],"by":[135],"+5.8%":[136],"1.4B":[139],"scale,":[140],"+10.9%":[142],"2.6B":[145],"scale.":[146],"Despite":[147],"being":[148],"trained":[149],"exclusively":[150],"mathematics,":[152],"also":[154],"transfers":[155],"effectively":[156],"non-mathematical":[158],"demonstrating":[161],"effectiveness":[163],"of":[164],"for":[168],"in":[171],"LoopLMs.":[172],"Code":[173],"is":[174],"available":[175],"https://github.com/jonwill8/RLTT.git.":[177]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-02-13T00:00:00"}