{"id":"https://openalex.org/W7151378172","doi":"https://doi.org/10.1109/icmla66185.2025.00140","title":"MOTIF: Modular Thinking via Reinforcement Fine-tuning in LLMs","display_name":"MOTIF: Modular Thinking via Reinforcement Fine-tuning in LLMs","publication_year":2025,"publication_date":"2025-12-03","ids":{"openalex":"https://openalex.org/W7151378172","doi":"https://doi.org/10.1109/icmla66185.2025.00140"},"language":null,"primary_location":{"id":"doi:10.1109/icmla66185.2025.00140","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icmla66185.2025.00140","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 International Conference on Machine Learning and Applications (ICMLA)","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5057409122","display_name":"Purbesh Mitra","orcid":"https://orcid.org/0000-0002-7243-9448"},"institutions":[{"id":"https://openalex.org/I66946132","display_name":"University of Maryland, College Park","ror":"https://ror.org/047s2c258","country_code":"US","type":"education","lineage":["https://openalex.org/I66946132"]}],"countries":["US"],"is_corresponding":true,"raw_author_name":"Purbesh Mitra","raw_affiliation_strings":["University of Maryland,Department of Electrical and Computer Engineering,College Park,MD,20742"],"affiliations":[{"raw_affiliation_string":"University of Maryland,Department of Electrical and Computer Engineering,College Park,MD,20742","institution_ids":["https://openalex.org/I66946132"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5133127430","display_name":"Sennur Ulukus","orcid":null},"institutions":[{"id":"https://openalex.org/I66946132","display_name":"University of Maryland, College Park","ror":"https://ror.org/047s2c258","country_code":"US","type":"education","lineage":["https://openalex.org/I66946132"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Sennur Ulukus","raw_affiliation_strings":["University of Maryland,Department of Electrical and Computer Engineering,College Park,MD,20742"],"affiliations":[{"raw_affiliation_string":"University of Maryland,Department of Electrical and Computer Engineering,College Park,MD,20742","institution_ids":["https://openalex.org/I66946132"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":["https://openalex.org/A5057409122"],"corresponding_institution_ids":["https://openalex.org/I66946132"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.7849649,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"936","last_page":"941"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10142","display_name":"Formal Methods in Verification","score":0.09600000083446503,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/T10142","display_name":"Formal Methods in Verification","score":0.09600000083446503,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/T10456","display_name":"Multi-Agent Systems and Negotiation","score":0.08540000021457672,"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/T10906","display_name":"AI-based Problem Solving and Planning","score":0.08179999887943268,"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/modular-design","display_name":"Modular design","score":0.5727999806404114},{"id":"https://openalex.org/keywords/reinforcement","display_name":"Reinforcement","score":0.5267999768257141},{"id":"https://openalex.org/keywords/action","display_name":"Action (physics)","score":0.39649999141693115},{"id":"https://openalex.org/keywords/control","display_name":"Control (management)","score":0.36469998955726624},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.25380000472068787}],"concepts":[{"id":"https://openalex.org/C101468663","wikidata":"https://www.wikidata.org/wiki/Q1620158","display_name":"Modular design","level":2,"score":0.5727999806404114},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.5267999768257141},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4253999888896942},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.39649999141693115},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.36469998955726624},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.3450999855995178},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.2856999933719635},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.2549999952316284},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.25380000472068787},{"id":"https://openalex.org/C2777614354","wikidata":"https://www.wikidata.org/wiki/Q55648394","display_name":"Modular construction","level":3,"score":0.25110000371932983}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/icmla66185.2025.00140","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icmla66185.2025.00140","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 International Conference on Machine Learning and Applications (ICMLA)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":5,"referenced_works":["https://openalex.org/W4411452954","https://openalex.org/W4416036482","https://openalex.org/W4417282850","https://openalex.org/W7133209886","https://openalex.org/W7133224126"],"related_works":[],"abstract_inverted_index":{"Recent":[0],"advancements":[1],"in":[2,69,119,166],"the":[3,26,51,60,81,124,134,167],"reasoning":[4,71],"capabilities":[5],"of":[6,45,63,76,83,178,184],"large":[7,74],"language":[8],"models":[9,27,189],"(LLMs)":[10],"show":[11,156],"that":[12],"employing":[13],"group":[14],"relative":[15],"policy":[16],"optimization":[17],"(GRPO)":[18],"algorithm":[19],"for":[20,33,115],"reinforcement":[21],"learning":[22],"(RL)":[23],"training":[24,113,165],"allows":[25],"to":[28,50,94,126],"use":[29],"more":[30],"thinking/reasoning":[31],"tokens":[32,46,118],"generating":[34,116],"better":[35],"responses.":[36],"However,":[37],"LLMs":[38],"can":[39],"generate":[40],"only":[41,176],"a":[42,67,90],"finite":[43],"amount":[44],"while":[47],"maintaining":[48],"attention":[49],"previously":[52],"generated":[53],"tokens.":[54,77],"This":[55],"limit,":[56],"also":[57],"known":[58],"as":[59],"context":[61,84,130],"size":[62],"an":[64,86,111],"LLM,":[65],"is":[66],"bottleneck":[68],"LLM":[70,87],"with":[72,128,175],"arbitrarily":[73],"number":[75],"To":[78],"think":[79,127],"beyond":[80],"limit":[82],"size,":[85],"must":[88],"employ":[89],"modular":[91],"thinking":[92,117],"strategy":[93],"reason":[95],"over":[96,161],"multiple":[97,120],"rounds.":[98],"In":[99],"this":[100,171],"work,":[101],"we":[102],"propose":[103],"MOTIF:":[104],"Modular":[105],"Thinking":[106],"via":[107,141],"Reinforcement":[108],"Fine-tuning":[109],"\u2013":[110],"RL":[112],"method":[114],"rounds,":[121],"effectively":[122],"allowing":[123],"model":[125,136],"additional":[129],"size.":[131],"We":[132],"trained":[133],"open-source":[135],"Qwen2.5-3B-Instruct":[137],"on":[138,149],"GSM8K":[139],"dataset":[140],"parameter":[142],"efficient":[143],"fine-tuning":[144],"and":[145,151,158,188,194],"tested":[146],"its":[147],"accuracy":[148],"MATH500":[150],"AIME2024":[152],"benchmarks.":[153,169],"Our":[154,186],"experiments":[155],"3.8%":[157],"3.3%":[159],"improvements":[160],"vanilla":[162],"GRPO":[163],"based":[164],"respective":[168],"Furthermore,":[170],"improvement":[172],"was":[173],"achieved":[174],"15%":[177],"samples,":[179],"thus":[180],"demonstrating":[181],"sample":[182],"efficiency":[183],"MOTIF.":[185],"code":[187],"are":[190],"available":[191],"at":[192],"https://github.com/purbeshmitra/MOTIF":[193],"https://huggingface.co/purbeshmitra/MOTIF,":[195],"respectively.":[196]},"counts_by_year":[],"updated_date":"2026-04-09T06:08:40.794217","created_date":"2026-04-08T00:00:00"}