{"id":"https://openalex.org/W7129102403","doi":"https://doi.org/10.48550/arxiv.2602.12556","title":"SD-MoE: Spectral Decomposition for Effective Expert Specialization","display_name":"SD-MoE: Spectral Decomposition for Effective Expert Specialization","publication_year":2026,"publication_date":"2026-02-13","ids":{"openalex":"https://openalex.org/W7129102403","doi":"https://doi.org/10.48550/arxiv.2602.12556"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2602.12556","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2602.12556","pdf_url":"https://arxiv.org/pdf/2602.12556","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2602.12556","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5124925822","display_name":"Ruijun Huang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Ruijun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5159742985","display_name":"Fang Dong","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dong, Fang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126165473","display_name":"Xin Zhang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Xin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124963427","display_name":"Hengjie Cao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cao, Hengjie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126164068","display_name":"Zhendong Huang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Zhendong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5049553013","display_name":"A. Q. Chen","orcid":"https://orcid.org/0009-0004-9712-0275","observed_orcids":["https://orcid.org/0009-0004-9712-0275"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Anrui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124909024","display_name":"Jixian Zhou","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Jixian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126099546","display_name":"Mengyi Chen","orcid":"https://orcid.org/0009-0007-9991-4409","observed_orcids":["https://orcid.org/0009-0007-9991-4409"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Mengyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126160503","display_name":"Yifeng Yang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang, Yifeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5029916784","display_name":"Mingzhi Dong","orcid":"https://orcid.org/0000-0002-8897-5931","observed_orcids":["https://orcid.org/0000-0002-8897-5931"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dong, Mingzhi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126111171","display_name":"Yujiang Wang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Yujiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5067557719","display_name":"Jinlong Hou","orcid":"https://orcid.org/0000-0002-7594-4714","observed_orcids":["https://orcid.org/0000-0002-7594-4714"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hou, Jinlong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126150430","display_name":"Qin Lv","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lv, Qin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5068224304","display_name":"Robert P. Dick","orcid":"https://orcid.org/0000-0001-5428-9530","observed_orcids":["https://orcid.org/0000-0001-5428-9530"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dick, Robert P.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126168198","display_name":"Yuan Cheng","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cheng, Yuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126098123","display_name":"Fan Yang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang, Fan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125065674","display_name":"Tun Lu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Tun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126116712","display_name":"Chun Zhang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Chun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5004722925","display_name":"Li Shang","orcid":"https://orcid.org/0000-0003-3944-7531","observed_orcids":["https://orcid.org/0000-0003-3944-7531"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shang, Li","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":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":0.7113999724388123,"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":0.7113999724388123,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.12359999865293503,"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":0.08749999850988388,"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":0.9660000205039978},{"id":"https://openalex.org/keywords/expert-specialization","display_name":"expert specialization","score":0.9660000205039978},{"id":"https://openalex.org/keywords/spectral-decomposition","display_name":"spectral decomposition","score":0.9670000076293945},{"id":"https://openalex.org/keywords/parameter-space","display_name":"parameter space","score":0.43700000643730164},{"id":"https://openalex.org/keywords/low-rank-structure","display_name":"low-rank structure","score":0.8289999961853027},{"id":"https://openalex.org/keywords/gating-mechanism","display_name":"gating mechanism","score":0.9010000228881836},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.8489999771118164},{"id":"https://openalex.org/keywords/spectral-decoupling","display_name":"spectral decoupling","score":0.574999988079071}],"concepts":[{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.6460999846458435},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6151999831199646},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.5712000131607056},{"id":"https://openalex.org/C204323151","wikidata":"https://www.wikidata.org/wiki/Q905424","display_name":"Range (aeronautics)","level":2,"score":0.47130000591278076},{"id":"https://openalex.org/C12362212","wikidata":"https://www.wikidata.org/wiki/Q728435","display_name":"Linear subspace","level":2,"score":0.4458000063896179},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.4293000102043152},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.39100000262260437},{"id":"https://openalex.org/C124681953","wikidata":"https://www.wikidata.org/wiki/Q339062","display_name":"Decomposition","level":2,"score":0.3896999955177307},{"id":"https://openalex.org/C2776207758","wikidata":"https://www.wikidata.org/wiki/Q5303302","display_name":"Downstream (manufacturing)","level":2,"score":0.36800000071525574},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.33169999718666077},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.32600000500679016},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.3077000081539154},{"id":"https://openalex.org/C58328972","wikidata":"https://www.wikidata.org/wiki/Q184609","display_name":"Expert system","level":2,"score":0.2854999899864197},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2793000042438507},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.27000001072883606},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.26809999346733093},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.2678999900817871},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.25870001316070557},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.25850000977516174},{"id":"https://openalex.org/C151201525","wikidata":"https://www.wikidata.org/wiki/Q177239","display_name":"Limit (mathematics)","level":2,"score":0.25780001282691956}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2602.12556","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2602.12556","pdf_url":"https://arxiv.org/pdf/2602.12556","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2602.12556","doi":"https://doi.org/10.48550/arxiv.2602.12556","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2602.12556","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2602.12556","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2602.12556","pdf_url":"https://arxiv.org/pdf/2602.12556","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4390833687","https://openalex.org/W7223216301","https://openalex.org/W4402954127","https://openalex.org/W4415964461","https://openalex.org/W4416512644","https://openalex.org/W7167718553","https://openalex.org/W4415274483","https://openalex.org/W7165061874","https://openalex.org/W7160986019","https://openalex.org/W7117940854"],"abstract_inverted_index":{"Mixture-of-Experts":[0],"(MoE)":[1],"architectures":[2],"scale":[3],"Large":[4],"Language":[5],"Models":[6],"via":[7],"expert":[8,17,125],"specialization":[9,18],"induced":[10],"by":[11,78],"conditional":[12],"computation.":[13],"In":[14,41],"practice,":[15],"however,":[16],"often":[19],"fails:":[20],"some":[21],"experts":[22,58],"become":[23],"functionally":[24],"similar,":[25],"while":[26],"others":[27],"functioning":[28],"as":[29],"de":[30],"facto":[31],"shared":[32],"experts,":[33,76],"limiting":[34,97],"the":[35,114],"effective":[36,124],"capacity":[37],"and":[38,52,85,111,131,146],"model":[39],"performance.":[40],"this":[42],"work,":[43],"we":[44,102],"analysis":[45],"from":[46],"a":[47,137],"spectral":[48,63,115],"perspective":[49],"on":[50],"parameter":[51,110],"gradient":[53,70,112],"spaces,":[54],"uncover":[55],"that":[56],"(1)":[57],"share":[59],"highly":[60],"overlapping":[61],"dominant":[62,69,94],"components":[64],"in":[65,82,113],"their":[66],"parameters,":[67],"(2)":[68],"subspaces":[71],"are":[72],"strongly":[73],"aligned":[74],"across":[75,120],"driven":[77],"ubiquitous":[79],"low-rank":[80],"structure":[81],"human":[83],"corpus,":[84],"(3)":[86],"gating":[87],"mechanisms":[88],"preferentially":[89],"route":[90],"inputs":[91],"along":[92],"these":[93],"directions,":[95],"further":[96],"specialization.":[98],"To":[99],"address":[100],"this,":[101],"propose":[103],"Spectral-Decoupled":[104],"MoE":[105,142],"(SD-MoE),":[106],"which":[107],"decomposes":[108],"both":[109],"space.":[116],"SD-MoE":[117],"improves":[118],"performance":[119],"downstream":[121],"tasks,":[122],"enables":[123],"specialization,":[126],"incurring":[127],"minimal":[128],"additional":[129],"computation,":[130],"can":[132],"be":[133],"seamlessly":[134],"integrated":[135],"into":[136],"wide":[138],"range":[139],"of":[140],"existing":[141],"architectures,":[143],"including":[144],"Qwen":[145],"DeepSeek.":[147]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-02-17T00:00:00"}