{"id":"https://openalex.org/W7214889453","doi":"https://doi.org/10.48550/arxiv.2609.36240","title":"Can Representation Learning Decouple from Loss Minimization? Polar Updates Have an Answer","display_name":"Can Representation Learning Decouple from Loss Minimization? Polar Updates Have an Answer","publication_year":2026,"publication_date":"2026-09-28","ids":{"openalex":"https://openalex.org/W7214889453","doi":"https://doi.org/10.48550/arxiv.2609.36240"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.36240","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.36240","pdf_url":"https://arxiv.org/pdf/2609.36240","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/2609.36240","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5073438119","display_name":"Akash Kumar","orcid":"https://orcid.org/0000-0002-4105-3681","observed_orcids":["https://orcid.org/0000-0002-4105-3681"]},"institutions":[{"id":"https://openalex.org/I36258959","display_name":"University of California San Diego","ror":"https://ror.org/0168r3w48","country_code":"US","type":"education","lineage":["https://openalex.org/I36258959"]}],"countries":["US"],"is_corresponding":true,"raw_author_name":"Kumar, Akash","raw_affiliation_strings":["Department of Computer Science and Engineering University of California San Diego"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science and Engineering University of California San Diego","institution_ids":["https://openalex.org/I36258959"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":["https://openalex.org/A5073438119"],"corresponding_institution_ids":["https://openalex.org/I36258959"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"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":0.975600004196167,"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":0.975600004196167,"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/T10320","display_name":"Neural Networks and Applications","score":0.014299999922513962,"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":0.006099999882280827,"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/representation-learning","display_name":"representation learning","score":0.9490000009536743},{"id":"https://openalex.org/keywords/teacher-student-learning","display_name":"teacher-student learning","score":0.6909999847412109},{"id":"https://openalex.org/keywords/feature-alignment","display_name":"feature alignment","score":0.5820000171661377},{"id":"https://openalex.org/keywords/mean-field-method","display_name":"mean-field analysis","score":0.6499999761581421},{"id":"https://openalex.org/keywords/relu-neural-networks","display_name":"ReLU networks","score":0.7179999947547913},{"id":"https://openalex.org/keywords/gelu","display_name":"GELU","score":0.9869999885559082},{"id":"https://openalex.org/keywords/resnet","display_name":"ResNet","score":0.6430000066757202},{"id":"https://openalex.org/keywords/momentum","display_name":"momentum","score":0.5199999809265137}],"concepts":[{"id":"https://openalex.org/C32834561","wikidata":"https://www.wikidata.org/wiki/Q660730","display_name":"Subspace topology","level":2,"score":0.633400022983551},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.5766000151634216},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5717999935150146},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.5297999978065491},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5177000164985657},{"id":"https://openalex.org/C162307627","wikidata":"https://www.wikidata.org/wiki/Q204833","display_name":"Enhanced Data Rates for GSM Evolution","level":2,"score":0.5177000164985657},{"id":"https://openalex.org/C2908647359","wikidata":"https://www.wikidata.org/wiki/Q2625603","display_name":"Population","level":2,"score":0.45660001039505005},{"id":"https://openalex.org/C12362212","wikidata":"https://www.wikidata.org/wiki/Q728435","display_name":"Linear subspace","level":2,"score":0.43560001254081726},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.41110000014305115},{"id":"https://openalex.org/C106487976","wikidata":"https://www.wikidata.org/wiki/Q685816","display_name":"Matrix (chemical analysis)","level":2,"score":0.37950000166893005},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.35670000314712524},{"id":"https://openalex.org/C2779227376","wikidata":"https://www.wikidata.org/wiki/Q6505497","display_name":"Layer (electronics)","level":2,"score":0.3537999987602234},{"id":"https://openalex.org/C151201525","wikidata":"https://www.wikidata.org/wiki/Q177239","display_name":"Limit (mathematics)","level":2,"score":0.33169999718666077},{"id":"https://openalex.org/C158693339","wikidata":"https://www.wikidata.org/wiki/Q190524","display_name":"Eigenvalues and eigenvectors","level":2,"score":0.3192000091075897},{"id":"https://openalex.org/C190839683","wikidata":"https://www.wikidata.org/wiki/Q2448197","display_name":"Train","level":2,"score":0.3151000142097473},{"id":"https://openalex.org/C164226766","wikidata":"https://www.wikidata.org/wiki/Q7293202","display_name":"Rank (graph theory)","level":2,"score":0.31279999017715454},{"id":"https://openalex.org/C155512373","wikidata":"https://www.wikidata.org/wiki/Q287450","display_name":"Residual","level":2,"score":0.30630001425743103},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.30489999055862427},{"id":"https://openalex.org/C2777735758","wikidata":"https://www.wikidata.org/wiki/Q817765","display_name":"Path (computing)","level":2,"score":0.2883000075817108},{"id":"https://openalex.org/C29705727","wikidata":"https://www.wikidata.org/wiki/Q294562","display_name":"Polar","level":2,"score":0.2865999937057495},{"id":"https://openalex.org/C59404180","wikidata":"https://www.wikidata.org/wiki/Q17013334","display_name":"Feature learning","level":2,"score":0.28519999980926514},{"id":"https://openalex.org/C60718061","wikidata":"https://www.wikidata.org/wiki/Q1414747","display_name":"Momentum (technical analysis)","level":2,"score":0.28450000286102295},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.2815000116825104},{"id":"https://openalex.org/C761482","wikidata":"https://www.wikidata.org/wiki/Q118093","display_name":"Transmission (telecommunications)","level":2,"score":0.27889999747276306},{"id":"https://openalex.org/C184720557","wikidata":"https://www.wikidata.org/wiki/Q7825049","display_name":"Topology (electrical circuits)","level":2,"score":0.27140000462532043},{"id":"https://openalex.org/C2988416141","wikidata":"https://www.wikidata.org/wiki/Q6031139","display_name":"Information loss","level":2,"score":0.2711000144481659},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.26989999413490295},{"id":"https://openalex.org/C62611344","wikidata":"https://www.wikidata.org/wiki/Q1062658","display_name":"Node (physics)","level":2,"score":0.25999999046325684},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.2540999948978424}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.36240","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.36240","pdf_url":"https://arxiv.org/pdf/2609.36240","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.2609.36240","doi":"https://doi.org/10.48550/arxiv.2609.36240","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.36240","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.36240","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.36240","pdf_url":"https://arxiv.org/pdf/2609.36240","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":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214889453.pdf","grobid_xml":"https://content.openalex.org/works/W7214889453.grobid-xml"},"referenced_works_count":12,"referenced_works":["https://openalex.org/W2053200087","https://openalex.org/W2088307891","https://openalex.org/W2770190057","https://openalex.org/W3009686669","https://openalex.org/W4313303785","https://openalex.org/W4391709409","https://openalex.org/W4392544739","https://openalex.org/W4407424967","https://openalex.org/W7133231470","https://openalex.org/W7162141601","https://openalex.org/W7164039050","https://openalex.org/W7165633675"],"related_works":["https://openalex.org/W7215033858","https://openalex.org/W7220470082","https://openalex.org/W7204047820","https://openalex.org/W4393072916","https://openalex.org/W7221130595","https://openalex.org/W7204248737","https://openalex.org/W7167635298","https://openalex.org/W7164009365","https://openalex.org/W2803589494","https://openalex.org/W7204266245"],"abstract_inverted_index":{"Does":[0],"representation":[1],"learning":[2,80],"stop":[3],"when":[4],"the":[5,26,34,54,62,67,74,111,115,122,131,150,159,163,174,196,200,230,238],"training":[6],"loss":[7,46,56,128,132,214,227],"stops":[8],"improving?":[9],"We":[10],"study":[11],"this":[12,224],"question":[13],"for":[14,50,179],"matrix":[15],"Muon,":[16],"whose":[17],"polar-normalised":[18],"updates":[19,208],"have":[20],"a":[21,96,127,210],"step":[22],"length":[23],"set":[24],"by":[25],"gradient's":[27],"rank":[28],"rather":[29],"than":[30],"its":[31],"norm.":[32],"Near":[33],"edge":[35],"of":[36,52,114,240],"stability,":[37],"full-batch":[38,205],"Muon":[39],"on":[40,169],"teacher-student":[41,79],"problems":[42],"enters":[43],"approximately":[44],"period-2":[45,164],"oscillations":[47],"that":[48,173],"persist":[49],"thousands":[51],"steps:":[53],"cycle-mean":[55,213],"stays":[57],"flat":[58,212],"or":[59,155],"rises,":[60],"yet":[61],"weights":[63],"keep":[64],"moving":[65],"and":[66,86,89,92,108,140,181,220,226,229,236],"learned":[68,175],"features":[69],"continue":[70],"to":[71,234,237],"align":[72],"with":[73,204,215],"teacher":[75,123,142,160],"subspace.":[76],"For":[77,95],"linear":[78],"toys,":[81],"we":[82,101,144],"derive":[83],"explicit":[84],"cycle":[85],"alignment":[87,147,186],"formulas":[88],"conditional":[90],"plateau":[91,225],"decay":[93],"bounds.":[94],"population":[97],"mean-field":[98],"ReLU":[99,193],"model,":[100],"prove":[102],"that,":[103],"under":[104],"stated":[105],"dimension,":[106],"initialisation":[107],"small-head":[109],"conditions,":[110],"leading":[112],"eigenspace":[113],"average":[116],"gradient":[117],"outer":[118],"product":[119],"(AGOP)":[120],"recovers":[121],"subspace":[124,161],"exactly":[125],"during":[126,162],"plateau,":[129],"before":[130],"later":[133],"drops.":[134],"In":[135,190],"all":[136],"33":[137],"ReLU,":[138],"GELU":[139],"SiLU":[141],"configurations":[143,171],"study,":[145],"direction-only":[146],"metrics":[148],"show":[149],"student":[151],"AGOP":[152,185],"aligned":[153],"with,":[154],"still":[156],"aligning":[157],"to,":[158],"oscillations;":[165],"projected":[166],"head":[167],"refitting":[168],"selected":[170],"shows":[172],"directions":[176],"are":[177],"useful":[178],"prediction,":[180],"further":[182],"measurements":[183],"distinguish":[184],"from":[187],"weight-mass":[188],"concentration.":[189],"deep":[191],"residual":[192],"students,":[194],"freezing":[195,219],"downstream":[197],"layers":[198],"while":[199],"first":[201],"layer":[202],"trains":[203],"exact":[206],"polar":[207],"recreates":[209],"nearly":[211],"improving":[216],"input-AGOP":[217],"alignment;":[218],"unfreezing":[221],"switch":[222],"between":[223],"decrease,":[228],"effect":[231],"is":[232],"sensitive":[233],"momentum":[235],"choice":[239],"orthogonaliser.":[241]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-01T00:00:00"}