{"id":"https://openalex.org/W7123963791","doi":"https://doi.org/10.1016/j.eswa.2026.131128","title":"Decoupling Mixture-of-experts Routing from Gradient Noise: A Framework for Structured Specialization and Soft Generalization Toward Robust and Efficient Inference","display_name":"Decoupling Mixture-of-experts Routing from Gradient Noise: A Framework for Structured Specialization and Soft Generalization Toward Robust and Efficient Inference","publication_year":2026,"publication_date":"2026-01-13","ids":{"openalex":"https://openalex.org/W7123963791","doi":"https://doi.org/10.1016/j.eswa.2026.131128"},"language":"en","primary_location":{"id":"doi:10.1016/j.eswa.2026.131128","doi":"https://doi.org/10.1016/j.eswa.2026.131128","is_oa":true,"landing_page_url":"https://doi.org/10.1016/j.eswa.2026.131128","pdf_url":null,"source":{"id":"https://openalex.org/S13144211","display_name":"Expert Systems with Applications","issn_l":"0957-4174","issn":["0957-4174","1873-6793"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["anvur-class-a","anvur-scientific","ccf-c","cwts-core","dongbi-c","doyens","fnege-3","jufo-2","ki-jl-2","norway-2","poland-200","russia-white-list-1","vabb-shw"],"host_organization":"https://openalex.org/P4310320990","host_organization_name":"Elsevier BV","host_organization_lineage":["https://openalex.org/P4310320990"],"host_organization_lineage_names":["Elsevier BV"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Expert Systems with Applications","raw_type":"journal-article"},"volume":null,"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://doi.org/10.1016/j.eswa.2026.131128","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5022530943","display_name":"Bakary Badjie","orcid":"https://orcid.org/0000-0001-8488-4955","observed_orcids":["https://orcid.org/0000-0001-8488-4955"]},"institutions":[{"id":"https://openalex.org/I141596103","display_name":"University of Lisbon","ror":"https://ror.org/01c27hj86","country_code":"PT","type":"education","lineage":["https://openalex.org/I141596103"]},{"id":"https://openalex.org/I4405257439","display_name":"Laborat\u00f3rio de Sistemas Inform\u00e1ticos de Grande Escala","ror":"https://ror.org/0597csq05","country_code":"PT","type":"facility","lineage":["https://openalex.org/I4405257439"]}],"countries":["PT"],"is_corresponding":true,"raw_author_name":"Bakary Badjie","raw_affiliation_strings":["LASIGE, Departamento de Inform\u00e1tica, Faculdade de Ci\u00eancias da Universidade de Lisboa, Campo Grande Edif\u00edcio C6, Lisboa, 1749-016, Portugal"],"raw_orcid":"https://orcid.org/0000-0001-8488-4955","affiliations":[{"raw_affiliation_string":"LASIGE, Departamento de Inform\u00e1tica, Faculdade de Ci\u00eancias da Universidade de Lisboa, Campo Grande Edif\u00edcio C6, Lisboa, 1749-016, Portugal","institution_ids":["https://openalex.org/I141596103","https://openalex.org/I4405257439"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5007948994","display_name":"Jos\u00e9 Cec\u00edlio","orcid":"https://orcid.org/0000-0002-5351-5580","observed_orcids":["https://orcid.org/0000-0002-5351-5580"]},"institutions":[{"id":"https://openalex.org/I141596103","display_name":"University of Lisbon","ror":"https://ror.org/01c27hj86","country_code":"PT","type":"education","lineage":["https://openalex.org/I141596103"]},{"id":"https://openalex.org/I4405257439","display_name":"Laborat\u00f3rio de Sistemas Inform\u00e1ticos de Grande Escala","ror":"https://ror.org/0597csq05","country_code":"PT","type":"facility","lineage":["https://openalex.org/I4405257439"]}],"countries":["PT"],"is_corresponding":false,"raw_author_name":"Jos\u00e9 Cec\u00edlio","raw_affiliation_strings":["LASIGE, Departamento de Inform\u00e1tica, Faculdade de Ci\u00eancias da Universidade de Lisboa, Campo Grande Edif\u00edcio C6, Lisboa, 1749-016, Portugal"],"raw_orcid":"https://orcid.org/0000-0002-5351-5580","affiliations":[{"raw_affiliation_string":"LASIGE, Departamento de Inform\u00e1tica, Faculdade de Ci\u00eancias da Universidade de Lisboa, Campo Grande Edif\u00edcio C6, Lisboa, 1749-016, Portugal","institution_ids":["https://openalex.org/I141596103","https://openalex.org/I4405257439"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5018449751","display_name":"Ant\u00f3nio Casimiro","orcid":"https://orcid.org/0000-0002-5522-5739","observed_orcids":["https://orcid.org/0000-0002-5522-5739"]},"institutions":[{"id":"https://openalex.org/I141596103","display_name":"University of Lisbon","ror":"https://ror.org/01c27hj86","country_code":"PT","type":"education","lineage":["https://openalex.org/I141596103"]},{"id":"https://openalex.org/I4405257439","display_name":"Laborat\u00f3rio de Sistemas Inform\u00e1ticos de Grande Escala","ror":"https://ror.org/0597csq05","country_code":"PT","type":"facility","lineage":["https://openalex.org/I4405257439"]}],"countries":["PT"],"is_corresponding":false,"raw_author_name":"Ant\u00f3nio Casimiro","raw_affiliation_strings":["LASIGE, Departamento de Inform\u00e1tica, Faculdade de Ci\u00eancias da Universidade de Lisboa, Campo Grande Edif\u00edcio C6, Lisboa, 1749-016, Portugal"],"raw_orcid":"https://orcid.org/0000-0002-5522-5739","affiliations":[{"raw_affiliation_string":"LASIGE, Departamento de Inform\u00e1tica, Faculdade de Ci\u00eancias da Universidade de Lisboa, Campo Grande Edif\u00edcio C6, Lisboa, 1749-016, Portugal","institution_ids":["https://openalex.org/I141596103","https://openalex.org/I4405257439"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":["https://openalex.org/A5022530943"],"corresponding_institution_ids":["https://openalex.org/I141596103","https://openalex.org/I4405257439"],"apc_list":{"value":3500,"currency":"USD","value_usd":3500},"apc_paid":{"value":3500,"currency":"USD","value_usd":3500},"fwci":3.1163,"has_fulltext":false,"cited_by_count":1,"citation_normalized_percentile":{"value":0.86037803,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":93,"max":96},"biblio":{"volume":"309","issue":null,"first_page":"131128","last_page":"131128"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9520999789237976,"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"}},"topics":[{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9520999789237976,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.041099999099969864,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.003800000064074993,"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/mixture-of-experts","display_name":"mixture of experts","score":0.9580000042915344},{"id":"https://openalex.org/keywords/expert-routing","display_name":"expert routing","score":0.8199999928474426},{"id":"https://openalex.org/keywords/gradient-noise","display_name":"gradient noise","score":0.7049999833106995},{"id":"https://openalex.org/keywords/expert-specialization","display_name":"expert specialization","score":0.7860000133514404},{"id":"https://openalex.org/keywords/feature-clustering","display_name":"feature clustering","score":0.8970000147819519},{"id":"https://openalex.org/keywords/siamese-network","display_name":"Siamese network","score":0.9139999747276306},{"id":"https://openalex.org/keywords/k-means-clustering","display_name":"K-means clustering","score":0.777999997138977},{"id":"https://openalex.org/keywords/k-nearest-neighbors","display_name":"k-nearest neighbors","score":0.7649999856948853},{"id":"https://openalex.org/keywords/bidirectional-training","display_name":"bidirectional training","score":0.7350000143051147},{"id":"https://openalex.org/keywords/inference-efficiency","display_name":"inference efficiency","score":0.6899999976158142},{"id":"https://openalex.org/keywords/interpretable-models","display_name":"interpretable models","score":0.5009999871253967}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8413000106811523},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.6658999919891357},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6079000234603882},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.5788999795913696},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.567300021648407},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.5097000002861023},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.5095000267028809},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.5031999945640564},{"id":"https://openalex.org/C49937458","wikidata":"https://www.wikidata.org/wiki/Q2599292","display_name":"Probabilistic logic","level":2,"score":0.4812000095844269},{"id":"https://openalex.org/C73555534","wikidata":"https://www.wikidata.org/wiki/Q622825","display_name":"Cluster analysis","level":2,"score":0.4348999857902527},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.42570000886917114},{"id":"https://openalex.org/C205606062","wikidata":"https://www.wikidata.org/wiki/Q5249645","display_name":"Decoupling (probability)","level":2,"score":0.4138000011444092},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.4036000072956085},{"id":"https://openalex.org/C2781181686","wikidata":"https://www.wikidata.org/wiki/Q4226068","display_name":"Coherence (philosophical gambling strategy)","level":2,"score":0.36570000648498535},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.3513999879360199},{"id":"https://openalex.org/C76522221","wikidata":"https://www.wikidata.org/wiki/Q5035396","display_name":"Multipath routing","level":5,"score":0.32829999923706055},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.32600000500679016},{"id":"https://openalex.org/C177818476","wikidata":"https://www.wikidata.org/wiki/Q12878103","display_name":"Hierarchical routing","level":5,"score":0.2976999878883362},{"id":"https://openalex.org/C136389625","wikidata":"https://www.wikidata.org/wiki/Q334384","display_name":"Supervised learning","level":3,"score":0.29019999504089355},{"id":"https://openalex.org/C105611402","wikidata":"https://www.wikidata.org/wiki/Q2976589","display_name":"Spectral clustering","level":3,"score":0.2897999882698059},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.28360000252723694},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.2831000089645386},{"id":"https://openalex.org/C59404180","wikidata":"https://www.wikidata.org/wiki/Q17013334","display_name":"Feature learning","level":2,"score":0.26809999346733093},{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.2556000053882599},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.2554999887943268},{"id":"https://openalex.org/C2780069185","wikidata":"https://www.wikidata.org/wiki/Q7977945","display_name":"Equivalence (formal languages)","level":2,"score":0.25519999861717224},{"id":"https://openalex.org/C58328972","wikidata":"https://www.wikidata.org/wiki/Q184609","display_name":"Expert system","level":2,"score":0.2513999938964844}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1016/j.eswa.2026.131128","doi":"https://doi.org/10.1016/j.eswa.2026.131128","is_oa":true,"landing_page_url":"https://doi.org/10.1016/j.eswa.2026.131128","pdf_url":null,"source":{"id":"https://openalex.org/S13144211","display_name":"Expert Systems with Applications","issn_l":"0957-4174","issn":["0957-4174","1873-6793"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["anvur-class-a","anvur-scientific","ccf-c","cwts-core","dongbi-c","doyens","fnege-3","jufo-2","ki-jl-2","norway-2","poland-200","russia-white-list-1","vabb-shw"],"host_organization":"https://openalex.org/P4310320990","host_organization_name":"Elsevier BV","host_organization_lineage":["https://openalex.org/P4310320990"],"host_organization_lineage_names":["Elsevier BV"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Expert Systems with Applications","raw_type":"journal-article"},{"id":"doi:10.2139/ssrn.5521537","doi":"https://doi.org/10.2139/ssrn.5521537","is_oa":true,"landing_page_url":"https://doi.org/10.2139/ssrn.5521537","pdf_url":null,"source":{"id":"https://openalex.org/S4210172589","display_name":"SSRN Electronic Journal","issn_l":"1556-5068","issn":["1556-5068"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I1318003438","host_organization_name":"RELX Group (Netherlands)","host_organization_lineage":["https://openalex.org/I1318003438"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":null,"raw_type":"posted-content"}],"best_oa_location":{"id":"doi:10.1016/j.eswa.2026.131128","doi":"https://doi.org/10.1016/j.eswa.2026.131128","is_oa":true,"landing_page_url":"https://doi.org/10.1016/j.eswa.2026.131128","pdf_url":null,"source":{"id":"https://openalex.org/S13144211","display_name":"Expert Systems with Applications","issn_l":"0957-4174","issn":["0957-4174","1873-6793"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["anvur-class-a","anvur-scientific","ccf-c","cwts-core","dongbi-c","doyens","fnege-3","jufo-2","ki-jl-2","norway-2","poland-200","russia-white-list-1","vabb-shw"],"host_organization":"https://openalex.org/P4310320990","host_organization_name":"Elsevier BV","host_organization_lineage":["https://openalex.org/P4310320990"],"host_organization_lineage_names":["Elsevier BV"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Expert Systems with Applications","raw_type":"journal-article"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[{"display_name":"Life below water","id":"https://metadata.un.org/sdg/14","score":0.5667991042137146}],"study_designs":[],"awards":[{"id":"https://openalex.org/G1927536394","display_name":"LASIGE Centro de Investiga\u00e7\u00e3o em Ci\u00eancia e Engenharia da Computa\u00e7\u00e3o","funder_award_id":"UID/00408/2025","funder_id":"https://openalex.org/F4320334779","funder_display_name":"Funda\u00e7\u00e3o para a Ci\u00eancia e a Tecnologia"},{"id":"https://openalex.org/G425899529","display_name":null,"funder_award_id":"2023.02832","funder_id":"https://openalex.org/F4320334779","funder_display_name":"Funda\u00e7\u00e3o para a Ci\u00eancia e a Tecnologia"}],"funders":[{"id":"https://openalex.org/F4320334779","display_name":"Funda\u00e7\u00e3o para a Ci\u00eancia e a Tecnologia","ror":"https://ror.org/00snfqn58"}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":5,"referenced_works":["https://openalex.org/W2025653905","https://openalex.org/W2068238590","https://openalex.org/W2150884987","https://openalex.org/W2983154683","https://openalex.org/W4323530522"],"related_works":[],"abstract_inverted_index":{"Mixture-of-Experts":[0],"(MoE)":[1],"models":[2],"enhance":[3],"deep":[4],"learning":[5],"scalability":[6],"by":[7,93],"activating":[8],"only":[9],"a":[10,64,75,85,94,131],"subset":[11],"of":[12,178],"specialized":[13],"subnetworks":[14],"per":[15],"input.":[16],"However,":[17],"conventional":[18],"MoEs":[19],"often":[20],"suffer":[21],"from":[22,71],"unstable":[23],"expert":[24,69,137,148,164],"specialization":[25,138,149,213],"and":[26,34,53,79,87,111,139,158,189,197,210,223],"biased":[27],"routing":[28,70,140,165],"due":[29],"to":[30,103,108,119,202],"entangled":[31],"gradient":[32,72],"updates":[33],"unstructured":[35],"input":[36],"handling.":[37],"To":[38],"address":[39],"these":[40,128],"problems,":[41],"this":[42],"paper":[43],"proposes":[44],"SEAS-GMoE":[45,83,193],"(":[46],"S":[47,54],"tructured":[48],"E":[49,62],"xpert":[50],"A":[51,122],"ssignment":[52],"upervised":[55],"G":[56],"ating":[57],"M":[58],"ixture":[59],"o":[60],"f":[61],"xperts),":[63],"hybrid":[65],"framework":[66],"that":[67,99,115,192],"decouples":[68],"noise":[73],"through":[74,130,168],"double-stage":[76],"feature":[77],"clustering":[78],"semantic":[80,101],"pseudo-labeling":[81],"mechanism.":[82],"employs":[84],"K-means":[86],"KNN-based":[88],"cluster":[89],"refinement":[90],"technique,":[91],"followed":[92],"Siamese":[95],"Neural":[96],"Network":[97],"(SNN)":[98],"assigns":[100],"pseudo-labels":[102],"clustered":[104],"features.":[105],"This":[106,174],"leads":[107],"generating":[109],"interpretable":[110],"semantically":[112],"coherent":[113],"clusters":[114],"are":[116],"explicitly":[117],"mapped":[118],"dedicated":[120],"experts.":[121],"supervised":[123],"managing":[124,170],"(gating)":[125],"network":[126],"learns":[127],"mappings":[129],"bidirectional":[132,144],"training":[133,145],"process,":[134,146],"reinforcing":[135],"both":[136],"reliability.":[141],"In":[142],"the":[143,151,169],"as":[147],"improves,":[150],"gating":[152],"supervision":[153],"signal":[154],"is":[155,166],"simultaneously":[156],"enhanced,":[157],"vice":[159],"versa.":[160],"During":[161],"inference,":[162],"soft":[163],"applied":[167],"network\u2019s":[171],"probabilistic":[172],"output.":[173],"enables":[175],"flexible":[176],"aggregation":[177],"expert(s)\u2019":[179],"predictions":[180],"on":[181,186],"new":[182],"inputs.":[183],"Experimental":[184],"results":[185,216],"GTSRB,":[187],"MNIST,":[188],"CIFAR-10":[190],"show":[191],"outperforms":[194],"reimplemented":[195],"V-MoE":[196],"dense":[198],"baselines,":[199],"achieving":[200],"up":[201],"4.1%":[203],"higher":[204],"accuracy,":[205],"60%":[206],"lower":[207],"inference":[208],"latency,":[209],"11%":[211],"greater":[212],"stability.":[214],"These":[215],"confirm":[217],"SEAS-GMoE\u2019s":[218],"effectiveness":[219],"for":[220],"robust,":[221],"interpretable,":[222],"efficient":[224],"inference.":[225]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-10-12T09:40:04.860316","created_date":"2026-01-14T00:00:00"}