{"id":"https://openalex.org/W7163062488","doi":"https://doi.org/10.48550/arxiv.2605.30992","title":"Eigenvectors of Experts are Training-free Non-collapsing Routers","display_name":"Eigenvectors of Experts are Training-free Non-collapsing Routers","publication_year":2026,"publication_date":"2026-05-29","ids":{"openalex":"https://openalex.org/W7163062488","doi":"https://doi.org/10.48550/arxiv.2605.30992"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.30992","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.30992","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.30992","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137521658","display_name":"Giang Do","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Do, Giang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137549946","display_name":"Hung Le","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Le, Hung","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137552203","display_name":"Truyen Tran","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tran, Truyen","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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.3131999969482422,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.3131999969482422,"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/T11273","display_name":"Advanced Graph Neural Networks","score":0.07909999787807465,"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/T11704","display_name":"Mobile Crowdsensing and Crowdsourcing","score":0.07540000230073929,"subfield":{"id":"https://openalex.org/subfields/1706","display_name":"Computer Science Applications"},"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/inference","display_name":"Inference","score":0.6320000290870667},{"id":"https://openalex.org/keywords/robustness","display_name":"Robustness (evolution)","score":0.6223999857902527},{"id":"https://openalex.org/keywords/encode","display_name":"ENCODE","score":0.551800012588501},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.37689998745918274},{"id":"https://openalex.org/keywords/focus","display_name":"Focus (optics)","score":0.3725999891757965},{"id":"https://openalex.org/keywords/generalization","display_name":"Generalization","score":0.3562999963760376},{"id":"https://openalex.org/keywords/structured-prediction","display_name":"Structured prediction","score":0.3400000035762787},{"id":"https://openalex.org/keywords/routing","display_name":"Routing (electronic design automation)","score":0.3231000006198883}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7714999914169312},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6320000290870667},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.6223999857902527},{"id":"https://openalex.org/C66746571","wikidata":"https://www.wikidata.org/wiki/Q1134833","display_name":"ENCODE","level":3,"score":0.551800012588501},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.47369998693466187},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.42829999327659607},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.37689998745918274},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.3725999891757965},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.3562999963760376},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.34459999203681946},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.3400000035762787},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.3231000006198883},{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.3100999891757965},{"id":"https://openalex.org/C22789450","wikidata":"https://www.wikidata.org/wiki/Q420904","display_name":"Singular value decomposition","level":2,"score":0.30799999833106995},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.30140000581741333},{"id":"https://openalex.org/C46743427","wikidata":"https://www.wikidata.org/wiki/Q1341685","display_name":"Inference engine","level":3,"score":0.2953000068664551},{"id":"https://openalex.org/C2522767166","wikidata":"https://www.wikidata.org/wiki/Q2374463","display_name":"Data science","level":1,"score":0.29019999504089355},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2851000130176544},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.26910001039505005},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.26840001344680786},{"id":"https://openalex.org/C2780598303","wikidata":"https://www.wikidata.org/wiki/Q65921492","display_name":"Flexibility (engineering)","level":2,"score":0.2676999866962433},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.2517000138759613},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.2500999867916107}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.30992","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.30992","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.30992","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.30992","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Sparse":[0],"Mixture":[1],"of":[2,10,23,114,151,185,194,201],"Experts":[3],"(SMoE)":[4],"architectures":[5],"improve":[6],"the":[7,39,57,83,105,112,152,157,180,199],"training":[8,31,64],"efficiency":[9],"Large":[11],"Language":[12],"Models":[13],"(LLMs)":[14],"by":[15,93],"routing":[16,129],"input":[17],"tokens":[18],"to":[19,123,127,155],"a":[20,142,191],"selected":[21],"subset":[22],"specialized":[24],"experts.":[25],"Despite":[26],"their":[27],"remarkable":[28],"success,":[29],"both":[30,94,173],"and":[32,73,96,109,144,160,169,175,183],"inference":[33],"in":[34],"SMoE":[35,89,107,140,204],"models":[36,108],"suffer":[37],"from":[38,65],"expert":[40,115,153],"collapse":[41,158],"issue":[42,84,159],"(Chi":[43],"et":[44],"al.,":[45],"2022),":[46],"which":[47,69],"degrades":[48],"model":[49,162,195],"performance.":[50,163],"Prior":[51],"studies":[52],"primarily":[53],"focus":[54],"on":[55,63,132],"improving":[56],"router;":[58],"however,":[59],"such":[60],"methods":[61],"rely":[62],"scratch":[66],"or":[67],"fine-tuning,":[68],"requires":[70],"high":[71],"computational":[72],"data-processing":[74],"costs.":[75],"Furthermore,":[76],"we":[77,103,135],"demonstrate":[78,179],"that,":[79],"despite":[80],"these":[81],"efforts,":[82],"persists":[85],"when":[86],"advancing":[87],"well-pretrained":[88],"models,":[90],"as":[91],"evidenced":[92],"theoretical":[95],"empirical":[97],"results.":[98],"To":[99],"fill":[100],"that":[101,111,147],"gap,":[102],"analyze":[104],"advanced":[106],"observe":[110],"eigenvectors":[113],"weight":[116],"matrices":[117],"encode":[118],"rich":[119],"semantic":[120],"information,":[121],"pointing":[122],"an":[124],"effective":[125,203],"alternative":[126],"conventional":[128],"strategies.":[130],"Building":[131],"this":[133],"insight,":[134],"propose":[136],"Singular":[137],"Value":[138],"Decomposition":[139],"(SSMoE),":[141],"novel":[143],"training-free":[145],"framework":[146],"leverages":[148],"spectral":[149],"properties":[150],"weights":[154],"address":[156],"enhance":[161],"Extensive":[164],"experiments":[165],"across":[166],"diverse":[167],"language":[168],"vision":[170],"tasks,":[171],"under":[172],"clean":[174],"corrupt":[176],"data":[177],"settings,":[178],"strong":[181],"generalization":[182],"robustness":[184],"SSMoE.":[186],"Our":[187,206],"findings":[188],"highlight":[189],"how":[190],"deeper":[192],"understanding":[193],"internals":[196],"can":[197],"guide":[198],"development":[200],"more":[202],"architectures.":[205],"implementation":[207],"is":[208],"publicly":[209],"available":[210],"at":[211],"https://github.com/giangdip2410/SSMoE.":[212]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-02T00:00:00"}