{"id":"https://openalex.org/W7140095155","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.323","title":"SD-E2: Semantic Exploration for Reasoning Under Token Budgets","display_name":"SD-E2: Semantic Exploration for Reasoning Under Token Budgets","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7140095155","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.323"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-eacl.323","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.323","pdf_url":"https://aclanthology.org/2026.findings-eacl.323.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EACL 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-eacl.323.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5130384731","display_name":"Kshitij Mishra","orcid":null},"institutions":[{"id":"https://openalex.org/I4210164862","display_name":"Artificial Intelligence in Medicine (Canada)","ror":"https://ror.org/05p590m36","country_code":"CA","type":"company","lineage":["https://openalex.org/I4210164862"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Kshitij Mishra","raw_affiliation_strings":["University of Artificial Intelligence"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Artificial Intelligence","institution_ids":["https://openalex.org/I4210164862"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5086633938","display_name":"Nils Lukas","orcid":"https://orcid.org/0009-0001-5891-9154"},"institutions":[{"id":"https://openalex.org/I4210164862","display_name":"Artificial Intelligence in Medicine (Canada)","ror":"https://ror.org/05p590m36","country_code":"CA","type":"company","lineage":["https://openalex.org/I4210164862"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Nils Lukas","raw_affiliation_strings":["University of Artificial Intelligence"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Artificial Intelligence","institution_ids":["https://openalex.org/I4210164862"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5030085635","display_name":"Salem Lahlou","orcid":null},"institutions":[{"id":"https://openalex.org/I4210164862","display_name":"Artificial Intelligence in Medicine (Canada)","ror":"https://ror.org/05p590m36","country_code":"CA","type":"company","lineage":["https://openalex.org/I4210164862"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Salem Lahlou","raw_affiliation_strings":["University of Artificial Intelligence"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Artificial Intelligence","institution_ids":["https://openalex.org/I4210164862"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I4210164862"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.24585062,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"6144","last_page":"6157"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11010","display_name":"Logic, Reasoning, and Knowledge","score":0.2395000010728836,"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/T11010","display_name":"Logic, Reasoning, and Knowledge","score":0.2395000010728836,"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/T11596","display_name":"Constraint Satisfaction and Optimization","score":0.10909999907016754,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T10215","display_name":"Semantic Web and Ontologies","score":0.06939999759197235,"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":0.3898000121116638},{"id":"https://openalex.org/keywords/semantics","display_name":"Semantics (computer science)","score":0.3831999897956848},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.27090001106262207},{"id":"https://openalex.org/keywords/semantic-network","display_name":"Semantic network","score":0.2606000006198883},{"id":"https://openalex.org/keywords/automated-reasoning","display_name":"Automated reasoning","score":0.25679999589920044}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6959999799728394},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.46650001406669617},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.3898000121116638},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.3831999897956848},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3433000147342682},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.27090001106262207},{"id":"https://openalex.org/C85407183","wikidata":"https://www.wikidata.org/wiki/Q1045785","display_name":"Semantic network","level":2,"score":0.2606000006198883},{"id":"https://openalex.org/C195344581","wikidata":"https://www.wikidata.org/wiki/Q2555318","display_name":"Automated reasoning","level":2,"score":0.25679999589920044},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2484000027179718},{"id":"https://openalex.org/C20162079","wikidata":"https://www.wikidata.org/wiki/Q1151406","display_name":"Case-based reasoning","level":2,"score":0.24220000207424164}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-eacl.323","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.323","pdf_url":"https://aclanthology.org/2026.findings-eacl.323.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EACL 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-eacl.323","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.323","pdf_url":"https://aclanthology.org/2026.findings-eacl.323.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7140095155.pdf","grobid_xml":"https://content.openalex.org/works/W7140095155.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Small":[0],"language":[1],"models":[2],"(SLMs)":[3],"struggle":[4],"with":[5,76],"complex":[6],"reasoning":[7,37,164],"because":[8],"exploration":[9,29],"is":[10,74],"expensive":[11],"under":[12],"tight":[13],"compute":[14],"budgets.We":[15],"introduce":[16],"Semantic":[17],"Diversity":[18],"-Exploration-Exploitation":[19],"(SD-E":[20],"2":[21,44,92,172],"),":[22],"a":[23,39,46,83,150,174],"reinforcement":[24],"learning":[25],"framework":[26],"that":[27,49,87,145],"makes":[28],"explicit":[30],"by":[31,103],"optimizing":[32],"semantic":[33,147],"diversity":[34,47,72],"in":[35,65,82,180],"generated":[36],"trajectories.Using":[38],"frozen":[40],"sentence-embedding":[41],"model,":[42],"SD-E":[43,91,171],"assigns":[45],"reward":[48,73],"captures":[50],"(i)":[51],"the":[52,94,133,163],"coverage":[53],"of":[54],"semantically":[55,115],"distinct":[56,116],"solution":[57,80],"strategies":[58,117],"and":[59,79,101,105,129],"(ii)":[60],"their":[61],"average":[62,113],"pairwise":[63],"dissimilarity":[64],"embedding":[66],"space,":[67],"rather":[68,167],"than":[69,168],"surface-form":[70],"novelty.This":[71],"combined":[75],"outcome":[77],"correctness":[78],"efficiency":[81,178],"z-score-normalized":[84],"multi-objective":[85],"objective":[86],"stabilizes":[88],"training.On":[89],"GSM8K,":[90],"surpasses":[93],"base":[95,128,141],"Qwen2.5-3B-Instructand":[96],"strong":[97],"GRPO":[98],"baselines":[99],"(GRPO-CFL":[100],"GRPO-CFEE)":[102],"+27.4,+5.3,":[104],"+1.5":[106],"percentage":[107],"points,":[108],"respectively,":[109],"while":[110],"discovering":[111],"on":[112,132],"9.8":[114],"per":[118],"question.We":[119],"further":[120],"improve":[121],"MedM-CQA":[122],"to":[123,177],"49.64%":[124],"vs":[125],"38.37":[126],"for":[127,155],"show":[130],"gains":[131,179],"harder":[134],"AIME":[135],"benchmark":[136],",":[137],"reaching":[138],"13.28%":[139],"vs.":[140],"6.74%.These":[142],"results":[143],"indicate":[144],"rewarding":[146],"novelty":[148],"yields":[149],"more":[151],"compute-efficient":[152],"exploration-exploitation":[153],"signal":[154],"training":[156],"reasoning-capable":[157],"SLMs.By":[158],"introducing":[159],"cognitive":[160],"adaptation":[161],"(adjusting":[162],"process":[165],"structure":[166],"per-token":[169],"computation),":[170],"offers":[173],"complementary":[175],"path":[176],"resource-constrained":[181],"models.":[182]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-03-24T00:00:00"}