{"id":"https://openalex.org/W7117108127","doi":"https://doi.org/10.48550/arxiv.2512.18452","title":"Secret mixtures of experts inside your LLM","display_name":"Secret mixtures of experts inside your LLM","publication_year":2025,"publication_date":"2025-12-20","ids":{"openalex":"https://openalex.org/W7117108127","doi":"https://doi.org/10.48550/arxiv.2512.18452"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2512.18452","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2512.18452","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2512.18452","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5091031925","display_name":"Enric Boix-Adser\u00e0","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Boix-Adsera, Enric","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":1,"corresponding_author_ids":["https://openalex.org/A5091031925"],"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.2865999937057495,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.2865999937057495,"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/T12611","display_name":"Neural Networks and Reservoir Computing","score":0.09520000219345093,"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/T11996","display_name":"Random lasers and scattering media","score":0.0649000033736229,"subfield":{"id":"https://openalex.org/subfields/3102","display_name":"Acoustics and Ultrasonics"},"field":{"id":"https://openalex.org/fields/31","display_name":"Physics and Astronomy"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.6448000073432922},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.6287000179290771},{"id":"https://openalex.org/keywords/autoencoder","display_name":"Autoencoder","score":0.5921000242233276},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.5350000262260437},{"id":"https://openalex.org/keywords/perceptron","display_name":"Perceptron","score":0.40470001101493835},{"id":"https://openalex.org/keywords/architecture","display_name":"Architecture","score":0.3946000039577484},{"id":"https://openalex.org/keywords/connection","display_name":"Connection (principal bundle)","score":0.3944000005722046},{"id":"https://openalex.org/keywords/models-of-neural-computation","display_name":"Models of neural computation","score":0.39010000228881836}],"concepts":[{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.6448000073432922},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6384999752044678},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6380000114440918},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.6287000179290771},{"id":"https://openalex.org/C101738243","wikidata":"https://www.wikidata.org/wiki/Q786435","display_name":"Autoencoder","level":3,"score":0.5921000242233276},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.5350000262260437},{"id":"https://openalex.org/C60908668","wikidata":"https://www.wikidata.org/wiki/Q690207","display_name":"Perceptron","level":3,"score":0.40470001101493835},{"id":"https://openalex.org/C123657996","wikidata":"https://www.wikidata.org/wiki/Q12271","display_name":"Architecture","level":2,"score":0.3946000039577484},{"id":"https://openalex.org/C13355873","wikidata":"https://www.wikidata.org/wiki/Q2920850","display_name":"Connection (principal bundle)","level":2,"score":0.3944000005722046},{"id":"https://openalex.org/C3832189","wikidata":"https://www.wikidata.org/wiki/Q8588916","display_name":"Models of neural computation","level":3,"score":0.39010000228881836},{"id":"https://openalex.org/C163716315","wikidata":"https://www.wikidata.org/wiki/Q901177","display_name":"Gaussian","level":2,"score":0.3817000091075897},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3709000051021576},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.33660000562667847},{"id":"https://openalex.org/C2984842247","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep neural networks","level":3,"score":0.320499986410141},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.29030001163482666},{"id":"https://openalex.org/C193415008","wikidata":"https://www.wikidata.org/wiki/Q639681","display_name":"Network architecture","level":2,"score":0.28040000796318054},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.2761000096797943},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.26269999146461487},{"id":"https://openalex.org/C61224824","wikidata":"https://www.wikidata.org/wiki/Q2260434","display_name":"Mixture model","level":2,"score":0.26010000705718994},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.25600001215934753},{"id":"https://openalex.org/C110121322","wikidata":"https://www.wikidata.org/wiki/Q865811","display_name":"Distribution (mathematics)","level":2,"score":0.2554999887943268}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2512.18452","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2512.18452","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2512.18452","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2512.18452","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/9","score":0.4117215871810913,"display_name":"Industry, innovation and infrastructure"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Despite":[0],"being":[1],"one":[2,15],"of":[3,16,21,32,69,126,152],"the":[4,9,17,22,40,96,104,124,150],"earliest":[5],"neural":[6,127],"network":[7,128],"layers,":[8],"Multilayer":[10],"Perceptron":[11],"(MLP)":[12],"is":[13,75],"arguably":[14],"least":[18],"understood":[19],"parts":[20],"transformer":[23],"architecture":[24,167],"due":[25],"to":[26,38],"its":[27],"dense":[28,44],"computation":[29,57],"and":[30,85,101,145],"lack":[31],"easy":[33],"visualization.":[34],"This":[35],"paper":[36],"seeks":[37],"understand":[39],"MLP":[41,141],"layers":[42,51,142],"in":[43,90,123,140],"LLM":[45],"models":[46,84],"by":[47,66],"hypothesizing":[48],"that":[49,60,103],"these":[50,109],"secretly":[52],"approximately":[53],"perform":[54],"a":[55,78,135],"sparse":[56],"--":[58,108],"namely,":[59],"they":[61],"can":[62],"be":[63],"well":[64],"approximated":[65],"sparsely-activating":[67],"Mixture":[68],"Experts":[70],"(MoE)":[71],"layers.":[72],"Our":[73,130],"hypothesis":[74,97],"based":[76,169],"on":[77,98,121,134,170],"novel":[79],"theoretical":[80],"connection":[81],"between":[82],"MoE":[83,166],"Sparse":[86],"Autoencoder":[87],"(SAE)":[88],"structure":[89,122],"activation":[91,105],"space.":[92],"We":[93],"empirically":[94],"validate":[95],"pretrained":[99],"LLMs,":[100,144],"demonstrate":[102],"distribution":[106,125],"matters":[107],"results":[110,131],"do":[111],"not":[112],"hold":[113],"for":[114,149,163],"Gaussian":[115],"data,":[116],"but":[117],"rather":[118],"rely":[119],"crucially":[120],"activations.":[129],"shine":[132],"light":[133],"general":[136],"principle":[137],"at":[138],"play":[139],"inside":[143],"give":[146],"an":[147],"explanation":[148],"effectiveness":[151],"modern":[153],"MoE-based":[154],"transformers.":[155],"Additionally,":[156],"our":[157],"experimental":[158],"explorations":[159],"suggest":[160],"new":[161],"directions":[162],"more":[164],"efficient":[165],"design":[168],"low-rank":[171],"routers.":[172]},"counts_by_year":[],"updated_date":"2025-12-24T23:14:05.333182","created_date":"2025-12-24T00:00:00"}