{"id":"https://openalex.org/W7163167451","doi":"https://doi.org/10.48550/arxiv.2606.00605","title":"Looped Transformers with Layer Normalization Provably Learn the Power Method","display_name":"Looped Transformers with Layer Normalization Provably Learn the Power Method","publication_year":2026,"publication_date":"2026-05-30","ids":{"openalex":"https://openalex.org/W7163167451","doi":"https://doi.org/10.48550/arxiv.2606.00605"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.00605","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.00605","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":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.00605","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137670706","display_name":"Lyumin Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Lyumin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137671404","display_name":"Chenyang Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Chenyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137620517","display_name":"Yuan Cao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cao, Yuan","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/T12611","display_name":"Neural Networks and Reservoir Computing","score":0.48429998755455017,"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/T12611","display_name":"Neural Networks and Reservoir Computing","score":0.48429998755455017,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.1160999983549118,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T11948","display_name":"Machine Learning in Materials Science","score":0.07620000094175339,"subfield":{"id":"https://openalex.org/subfields/2505","display_name":"Materials Chemistry"},"field":{"id":"https://openalex.org/fields/25","display_name":"Materials Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.6873999834060669},{"id":"https://openalex.org/keywords/normalization","display_name":"Normalization (sociology)","score":0.48010000586509705},{"id":"https://openalex.org/keywords/testbed","display_name":"Testbed","score":0.44670000672340393},{"id":"https://openalex.org/keywords/gradient-descent","display_name":"Gradient descent","score":0.4180999994277954},{"id":"https://openalex.org/keywords/principal-component-analysis","display_name":"Principal component analysis","score":0.4002000093460083},{"id":"https://openalex.org/keywords/current-transformer","display_name":"Current transformer","score":0.31949999928474426}],"concepts":[{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6873999834060669},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4952999949455261},{"id":"https://openalex.org/C136886441","wikidata":"https://www.wikidata.org/wiki/Q926129","display_name":"Normalization (sociology)","level":2,"score":0.48010000586509705},{"id":"https://openalex.org/C31395832","wikidata":"https://www.wikidata.org/wiki/Q1318674","display_name":"Testbed","level":2,"score":0.44670000672340393},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.4180999994277954},{"id":"https://openalex.org/C27438332","wikidata":"https://www.wikidata.org/wiki/Q2873","display_name":"Principal component analysis","level":2,"score":0.4002000093460083},{"id":"https://openalex.org/C24326235","wikidata":"https://www.wikidata.org/wiki/Q126095","display_name":"Electronic engineering","level":1,"score":0.3882000148296356},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.32010000944137573},{"id":"https://openalex.org/C133770746","wikidata":"https://www.wikidata.org/wiki/Q856535","display_name":"Current transformer","level":4,"score":0.31949999928474426},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3109000027179718},{"id":"https://openalex.org/C89227174","wikidata":"https://www.wikidata.org/wiki/Q2388981","display_name":"Electric power system","level":3,"score":0.3050999939441681},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.30140000581741333},{"id":"https://openalex.org/C117323899","wikidata":"https://www.wikidata.org/wiki/Q1907989","display_name":"Distribution transformer","level":4,"score":0.28450000286102295},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.27790001034736633},{"id":"https://openalex.org/C47446073","wikidata":"https://www.wikidata.org/wiki/Q5165890","display_name":"Control theory (sociology)","level":3,"score":0.27549999952316284},{"id":"https://openalex.org/C133731056","wikidata":"https://www.wikidata.org/wiki/Q4917288","display_name":"Control engineering","level":1,"score":0.2718000113964081}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.00605","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.00605","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":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.00605","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.00605","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":"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":{"Transformers":[0],"have":[1],"achieved":[2,142],"remarkable":[3],"success":[4],"across":[5],"a":[6,12,61,76,88,159,175,194],"wide":[7],"range":[8],"of":[9,15,20,36,48,69,131,212,216,229],"applications,":[10],"and":[11,29,165,218,223],"growing":[13],"body":[14],"work":[16],"suggests":[17],"that":[18,75,90,151],"part":[19],"their":[21,25],"strength":[22],"comes":[23],"from":[24,172],"ability":[26],"to":[27,87,118,193,205],"learn":[28,39,181],"execute":[30],"algorithmic":[31],"procedures.":[32],"However,":[33],"our":[34,206],"understanding":[35,65],"how":[37],"transformers":[38,70,133,163,220],"such":[40],"algorithms":[41],"remains":[42],"limited,":[43],"especially":[44],"in":[45,139,198,231],"the":[46,66,92,104,120,153,182,186,208,213,227],"presence":[47],"layer":[49,98],"normalization":[50],"(LN).":[51],"In":[52],"this":[53],"work,":[54],"we":[55],"study":[56],"principal":[57,110,199],"component":[58,111,200],"prediction":[59,137],"as":[60],"concrete":[62,160],"testbed":[63],"for":[64,109],"training":[67,214],"dynamics":[68,215],"with":[71,80,95,134,164,169,189,221],"LN.":[72],"We":[73,156],"prove":[74],"looped":[77,132,217],"linear":[78],"transformer":[79,176,188,232],"LN,":[81,222],"trained":[82,107],"by":[83,143],"gradient":[84,147],"descent,":[85],"converges":[86],"solution":[89],"implements":[91],"power":[93,101,121,154,173,183],"method,":[94,184],"each":[96],"self-attention":[97],"performing":[99],"one":[100,150],"iteration.":[102],"Notably,":[103],"model":[105],"is":[106],"only":[108],"prediction,":[112],"rather":[113],"than":[114],"being":[115],"explicitly":[116],"supervised":[117],"implement":[119],"method.":[122,155],"Our":[123,202],"finding":[124],"thus":[125],"reveals":[126],"an":[127],"\"algorithmic":[128],"implicit":[129],"bias\"":[130],"LN:":[135,167],"principal-component":[136],"can":[138],"principle":[140],"be":[141],"many":[144],"mechanisms,":[145],"yet":[146],"descent":[148],"selects":[149],"realizes":[152],"further":[157],"provide":[158],"comparison":[161],"between":[162],"without":[166,177],"even":[168],"layerwise":[170],"guidance":[171],"iterations,":[174],"LN":[178,190,230],"cannot":[179],"exactly":[180],"whereas":[185],"corresponding":[187],"can,":[191],"leading":[192],"provable":[195],"performance":[196],"gap":[197],"prediction.":[201],"results":[203],"provide,":[204],"knowledge,":[207],"first":[209],"theoretical":[210],"analysis":[211],"single-layer":[219],"shed":[224],"light":[225],"on":[226],"role":[228],"models.":[233]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-03T00:00:00"}