{"id":"https://openalex.org/W7220735899","doi":"https://doi.org/10.48550/arxiv.2610.04140","title":"ExpertMuon-Compass: Alignment-Guided Step Sizes for Mixture-of-Experts Training","display_name":"ExpertMuon-Compass: Alignment-Guided Step Sizes for Mixture-of-Experts Training","publication_year":2026,"publication_date":"2026-10-02","ids":{"openalex":"https://openalex.org/W7220735899","doi":"https://doi.org/10.48550/arxiv.2610.04140"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.04140","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.04140","pdf_url":"https://arxiv.org/pdf/2610.04140","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","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2610.04140","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5055597754","display_name":"Omatharv Bharat Vaidya","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vaidya, Omatharv Bharat","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133573243","display_name":"Ashwin Vinod","orcid":"https://orcid.org/0000-0001-5671-115X","observed_orcids":["https://orcid.org/0000-0001-5671-115X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vinod, Ashwin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5092950223","display_name":"Pedram Akbarian","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Akbarian, Pedram","raw_affiliation_strings":["Equal contribution"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Equal contribution","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5122202487","display_name":"Aditya Sai Ellendula","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ellendula, Aditya Sai","raw_affiliation_strings":["Equal contribution"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Equal contribution","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5026867670","display_name":"Connor Jerzak","orcid":"https://orcid.org/0000-0003-1914-8905","observed_orcids":["https://orcid.org/0000-0003-1914-8905"]},"institutions":[{"id":"https://openalex.org/I86519309","display_name":"The University of Texas at Austin","ror":"https://ror.org/00hj54h04","country_code":"US","type":"education","lineage":["https://openalex.org/I86519309"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jerzak, Connor T.","raw_affiliation_strings":["University of Texas at Austin"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Texas at Austin","institution_ids":["https://openalex.org/I86519309"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5127869926","display_name":"Nhat Ho","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I86519309","display_name":"The University of Texas at Austin","ror":"https://ror.org/00hj54h04","country_code":"US","type":"education","lineage":["https://openalex.org/I86519309"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ho, Nhat","raw_affiliation_strings":["University of Texas at Austin"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Texas at Austin","institution_ids":["https://openalex.org/I86519309"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":2,"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":0.7289999723434448,"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":0.7289999723434448,"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/T10028","display_name":"Topic Modeling","score":0.1289999932050705,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.0560000017285347,"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":0.9459999799728394},{"id":"https://openalex.org/keywords/muon-optimizer","display_name":"Muon optimizer","score":0.7689999938011169},{"id":"https://openalex.org/keywords/gradient-alignment","display_name":"gradient alignment","score":0.6620000004768372},{"id":"https://openalex.org/keywords/nesterov-momentum","display_name":"Nesterov momentum","score":0.7670000195503235},{"id":"https://openalex.org/keywords/weight-decay","display_name":"weight decay","score":0.6000000238418579},{"id":"https://openalex.org/keywords/optimizer-design","display_name":"optimizer design","score":0.41600000858306885},{"id":"https://openalex.org/keywords/multilingual-language-models","display_name":"multilingual language models","score":0.5630000233650208},{"id":"https://openalex.org/keywords/perturbation-bounds","display_name":"perturbation bounds","score":0.7670000195503235}],"concepts":[{"id":"https://openalex.org/C2778361833","wikidata":"https://www.wikidata.org/wiki/Q34735","display_name":"Compass","level":2,"score":0.5716000199317932},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5591999888420105},{"id":"https://openalex.org/C178009071","wikidata":"https://www.wikidata.org/wiki/Q93344","display_name":"Trigonometric functions","level":2,"score":0.42669999599456787},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4083999991416931},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.40689998865127563},{"id":"https://openalex.org/C58328972","wikidata":"https://www.wikidata.org/wiki/Q184609","display_name":"Expert system","level":2,"score":0.39910000562667847},{"id":"https://openalex.org/C57691317","wikidata":"https://www.wikidata.org/wiki/Q1289248","display_name":"Scalar (mathematics)","level":2,"score":0.38850000500679016},{"id":"https://openalex.org/C2775896111","wikidata":"https://www.wikidata.org/wiki/Q642560","display_name":"Router","level":2,"score":0.3702999949455261},{"id":"https://openalex.org/C34736171","wikidata":"https://www.wikidata.org/wiki/Q918333","display_name":"Preprocessor","level":2,"score":0.367900013923645},{"id":"https://openalex.org/C60718061","wikidata":"https://www.wikidata.org/wiki/Q1414747","display_name":"Momentum (technical analysis)","level":2,"score":0.367900013923645},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.3422999978065491},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.3384999930858612},{"id":"https://openalex.org/C29001434","wikidata":"https://www.wikidata.org/wiki/Q8084","display_name":"Trigonometry","level":2,"score":0.3312999904155731},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.32760000228881836},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.3149999976158142},{"id":"https://openalex.org/C77553402","wikidata":"https://www.wikidata.org/wiki/Q13222579","display_name":"Upper and lower bounds","level":2,"score":0.3018999993801117},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.29589998722076416},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2913999855518341},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.27480000257492065},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2728999853134155},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2709999978542328},{"id":"https://openalex.org/C3020580240","wikidata":"https://www.wikidata.org/wiki/Q663272","display_name":"Expert opinion","level":2,"score":0.26969999074935913},{"id":"https://openalex.org/C95623464","wikidata":"https://www.wikidata.org/wiki/Q1096149","display_name":"Classifier (UML)","level":2,"score":0.2696000039577484}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.04140","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.04140","pdf_url":"https://arxiv.org/pdf/2610.04140","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.2610.04140","doi":"https://doi.org/10.48550/arxiv.2610.04140","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.04140","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:2610.04140","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.04140","pdf_url":"https://arxiv.org/pdf/2610.04140","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":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7220735899.pdf","grobid_xml":"https://content.openalex.org/works/W7220735899.grobid-xml"},"referenced_works_count":12,"referenced_works":["https://openalex.org/W1483804921","https://openalex.org/W2581624817","https://openalex.org/W4394780940","https://openalex.org/W4402705925","https://openalex.org/W4405033002","https://openalex.org/W4406032137","https://openalex.org/W4414846785","https://openalex.org/W4414977079","https://openalex.org/W4415798387","https://openalex.org/W4416617024","https://openalex.org/W7130325503","https://openalex.org/W7131620822"],"related_works":["https://openalex.org/W7204770309","https://openalex.org/W7166831075","https://openalex.org/W4389524291","https://openalex.org/W7155381418","https://openalex.org/W7162767780","https://openalex.org/W4399353875","https://openalex.org/W7166845002","https://openalex.org/W7160986019","https://openalex.org/W7211902890","https://openalex.org/W7223216301"],"abstract_inverted_index":{"Mixture-of-experts":[0],"(MoE)":[1],"language":[2],"models":[3],"send":[4],"each":[5,12,72,188],"token":[6],"to":[7,42,158,167,219],"a":[8,17,30,173,200,226],"few":[9],"experts,":[10],"so":[11],"expert":[13,43,73,189,210],"is":[14,54,179],"trained":[15],"on":[16,133,139],"different":[18],"part":[19,25],"of":[20,37,45,71,111,129,199],"the":[21,39,46,68,81,84,92,96,106,112,122,126,161,170,180,184,197,209,215,230],"data,":[22],"and":[23,88,114,125,151,214],"this":[24],"changes":[26],"during":[27,191],"training.":[28],"With":[29,207],"shared":[31],"learning":[32],"rate,":[33],"Muon":[34,69],"applies":[35],"updates":[36],"roughly":[38],"same":[40,47,93,171],"size":[41],"matrices":[44,141],"shape,":[48],"even":[49],"when":[50,183,196],"an":[51],"expert's":[52,85],"update":[53,87,113,123],"poorly":[55],"aligned":[56],"with":[57,91,136,154],"its":[58,89,100,165],"current":[59],"gradient.":[60],"We":[61,223],"here":[62],"propose":[63],"ExpertMuon-Compass":[64],"(Compass),":[65],"which":[66],"multiplies":[67],"step":[70],"by":[74,187],"two":[75,231],"factors.":[76,232],"A":[77,102],"family":[78],"factor":[79],"compares":[80],"cosine":[82,94],"between":[83,108],"orthogonalized":[86],"gradient":[90,115],"for":[95,193,229],"other":[97,152],"experts":[98,220],"in":[99,160,204],"layer.":[101],"scalar":[103],"radius":[104],"aggregates":[105],"alignment":[107],"corresponding":[109],"rows":[110],"into":[116],"one":[117],"step-length":[118],"multiplier.":[119],"Compass":[120,135,178],"keeps":[121],"direction":[124],"momentum":[127,138],"buffer":[128],"Muon.":[130],"In":[131],"pretraining":[132],"FineWeb-Edu,":[134],"Nesterov":[137],"all":[140],"performs":[142],"as":[143,145,195],"well":[144],"or":[146,172],"better":[147],"than":[148,176],"Muon,":[149],"NorMuon,":[150],"optimizers,":[153],"weight":[155],"decay":[156],"matched":[157],"NorMuon":[159,168],"longer":[162],"runs.":[163],"Adding":[164],"factors":[166],"gives":[169],"lower":[174],"loss":[175],"NorMuon.":[177],"most":[181],"effective":[182],"data":[185],"seen":[186],"varies":[190],"training,":[192],"example,":[194],"languages":[198],"multilingual":[201],"corpus":[202],"arrive":[203],"separate":[205],"blocks.":[206],"Compass,":[208],"load":[211],"stays":[212],"balanced,":[213],"router":[216],"assigns":[217],"tokens":[218],"more":[221],"decisively.":[222],"also":[224],"prove":[225],"perturbation":[227],"bound":[228]},"counts_by_year":[{"year":2026,"cited_by_count":2}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-07T00:00:00"}