{"id":"https://openalex.org/W7166847521","doi":"https://doi.org/10.18653/v1/2026.acl-long.1675","title":"Union-of-Experts: Neurons in Mixture-of-Experts are Secretly Routers","display_name":"Union-of-Experts: Neurons in Mixture-of-Experts are Secretly Routers","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166847521","doi":"https://doi.org/10.18653/v1/2026.acl-long.1675"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.acl-long.1675","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.1675","pdf_url":"https://aclanthology.org/2026.acl-long.1675.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.acl-long.1675.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5139749664","display_name":"Songhao Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Songhao Wu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139756526","display_name":"Ang Lv","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ang Lv","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139738599","display_name":"Ruobing Xie","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ruobing Xie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125052814","display_name":"Samm Sun","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Samm Sun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139819368","display_name":"Di Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Di Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139849867","display_name":"Rui Yan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rui Yan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139774595","display_name":"Yankai Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yankai Lin","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":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.86073826,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"36193","last_page":"36206"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10581","display_name":"Neural dynamics and brain function","score":0.12530000507831573,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},"topics":[{"id":"https://openalex.org/T10581","display_name":"Neural dynamics and brain function","score":0.12530000507831573,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},{"id":"https://openalex.org/T11094","display_name":"Face Recognition and Perception","score":0.06459999829530716,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},{"id":"https://openalex.org/T10448","display_name":"Memory and Neural Mechanisms","score":0.05620000138878822,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/nervous-system","display_name":"Nervous system","score":0.3068000078201294},{"id":"https://openalex.org/keywords/mesoderm","display_name":"Mesoderm","score":0.23489999771118164},{"id":"https://openalex.org/keywords/central-nervous-system","display_name":"Central nervous system","score":0.2345000058412552},{"id":"https://openalex.org/keywords/signal-transduction","display_name":"Signal transduction","score":0.20839999616146088}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.3652999997138977},{"id":"https://openalex.org/C169760540","wikidata":"https://www.wikidata.org/wiki/Q207011","display_name":"Neuroscience","level":1,"score":0.33410000801086426},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.32019999623298645},{"id":"https://openalex.org/C545706735","wikidata":"https://www.wikidata.org/wiki/Q9404","display_name":"Nervous system","level":2,"score":0.3068000078201294},{"id":"https://openalex.org/C71924100","wikidata":"https://www.wikidata.org/wiki/Q11190","display_name":"Medicine","level":0,"score":0.30149999260902405},{"id":"https://openalex.org/C2780915607","wikidata":"https://www.wikidata.org/wiki/Q323438","display_name":"Mesoderm","level":4,"score":0.23489999771118164},{"id":"https://openalex.org/C529278444","wikidata":"https://www.wikidata.org/wiki/Q47273","display_name":"Central nervous system","level":2,"score":0.2345000058412552},{"id":"https://openalex.org/C95444343","wikidata":"https://www.wikidata.org/wiki/Q7141","display_name":"Cell biology","level":1,"score":0.21930000185966492},{"id":"https://openalex.org/C62478195","wikidata":"https://www.wikidata.org/wiki/Q828130","display_name":"Signal transduction","level":2,"score":0.20839999616146088},{"id":"https://openalex.org/C185592680","wikidata":"https://www.wikidata.org/wiki/Q2329","display_name":"Chemistry","level":0,"score":0.2069000005722046}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.acl-long.1675","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.1675","pdf_url":"https://aclanthology.org/2026.acl-long.1675.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.acl-long.1675","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.1675","pdf_url":"https://aclanthology.org/2026.acl-long.1675.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[{"score":0.5737529397010803,"id":"https://metadata.un.org/sdg/2","display_name":"Zero hunger"}],"awards":[{"id":"https://openalex.org/G3114819178","display_name":null,"funder_award_id":"2023QNRC001","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G3248486410","display_name":null,"funder_award_id":"U2436209","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"funders":[{"id":"https://openalex.org/F4320316083","display_name":"Tencent","ror":"https://ror.org/00hhjss72"},{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166847521.pdf","grobid_xml":"https://content.openalex.org/works/W7166847521.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Mixture-of-Experts":[0],"(MoE)":[1],"models":[2,127],"rely":[3],"on":[4],"an":[5,37],"external":[6],"router":[7],"to":[8,11,25,49,130],"assign":[9],"tokens":[10,65],"experts.This":[12],"design":[13],"inherently":[14],"separates":[15],"the":[16,68,83,100,117,145,156],"routing":[17,60,75,93,146],"decision":[18],"from":[19,87],"each":[20,57,78],"expert's":[21,79],"internal":[22],"capabilities,":[23],"leading":[24],"suboptimal":[26],"performance.In":[27],"this":[28,32],"work,":[29],"we":[30,90],"address":[31],"limitation":[33],"with":[34,77,128,139],"Union-of-Experts":[35],"(UoE),":[36],"MoE":[38,159],"variant":[39],"that":[40,134],"performs":[41],"\"expertautonomous":[42],"routing\".The":[43],"core":[44],"mechanism":[45],"of":[46,54,71,85,144,158],"UoE":[47,126],"is":[48],"pre-designate":[50],"a":[51,107],"minute":[52],"fraction":[53],"neurons":[55,147],"within":[56],"expert":[58,111],"as":[59,106],"neurons.Experts":[61],"autonomously":[62],"select":[63],"relevant":[64],"by":[66],"comparing":[67],"activation":[69],"intensity":[70],"these":[72],"neurons,":[73],"aligning":[74],"decisions":[76],"functional":[80],"profile.To":[81],"prevent":[82],"waste":[84],"activations":[86],"unselected":[88],"experts,":[89,119],"aggregate":[91],"all":[92],"neuron":[94],"outputs":[95],"and":[96,120,154],"sum":[97],"them":[98],"into":[99,151],"final":[101],"layer":[102],"output.This":[103],"aggregation":[104],"acts":[105],"novel":[108],"virtual":[109],"shared":[110],"whose":[112],"parameters":[113],"are":[114],"distributed":[115],"across":[116],"individual":[118],"improves":[121],"overall":[122],"parameter":[123],"efficiency.We":[124],"pre-train":[125],"up":[129],"3B":[131],"parameters,":[132],"demonstrating":[133],"they":[135],"outperform":[136],"traditional":[137],"MoEs":[138],"matched":[140],"efficiency.Furthermore,":[141],"our":[142],"analysis":[143],"provides":[148],"valuable":[149],"insights":[150],"expert-autonomous":[152],"selection":[153],"advances":[155],"understanding":[157],"routing.":[160]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}