{"id":"https://openalex.org/W7133201205","doi":"https://doi.org/10.52202/075280-0521","title":"What can a Single Attention Layer Learn? A Study Through the Random Features Lens","display_name":"What can a Single Attention Layer Learn? A Study Through the Random Features Lens","publication_year":2023,"publication_date":"2023-01-01","ids":{"openalex":"https://openalex.org/W7133201205","doi":"https://doi.org/10.52202/075280-0521"},"language":"en","primary_location":{"id":"doi:10.52202/075280-0521","doi":"https://doi.org/10.52202/075280-0521","is_oa":false,"landing_page_url":"https://doi.org/10.52202/075280-0521","pdf_url":null,"source":{"id":"https://openalex.org/S4363606243","display_name":"neural information processing systems","issn_l":"1049-5258","issn":["1049-5258"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-2"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Advances in Neural Information Processing Systems 36","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","is_traditional":true,"indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2307.11353","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5114137722","display_name":"Hengyu Fu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hengyu Fu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5121251544","display_name":"Tianyu Guo","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tianyu Guo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5146852008","display_name":"Yu Bai","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu Bai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5127794508","display_name":"Song Mei","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Song Mei","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":2.1535,"has_fulltext":true,"cited_by_count":19,"citation_normalized_percentile":{"value":0.91670015,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":96,"max":98},"biblio":{"volume":null,"issue":null,"first_page":"11912","last_page":"11951"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.6904000043869019,"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":0.6904000043869019,"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":0.17980000376701355,"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":0.02810000069439411,"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/multi-head-attention","display_name":"multi-head attention","score":0.8149999976158142},{"id":"https://openalex.org/keywords/random-features","display_name":"random features","score":0.6729999780654907},{"id":"https://openalex.org/keywords/transformer","display_name":"transformer models","score":0.6819999814033508},{"id":"https://openalex.org/keywords/excess-risk-bounds","display_name":"excess risk bounds","score":0.8489999771118164},{"id":"https://openalex.org/keywords/sample-complexity","display_name":"sample complexity","score":0.8529999852180481},{"id":"https://openalex.org/keywords/permutation-invariance","display_name":"permutation invariance","score":0.9399999976158142},{"id":"https://openalex.org/keywords/generalization","display_name":"generalization","score":0.625}],"concepts":[{"id":"https://openalex.org/C120665830","wikidata":"https://www.wikidata.org/wiki/Q14620","display_name":"Optics","level":1,"score":0.5163000226020813},{"id":"https://openalex.org/C2779227376","wikidata":"https://www.wikidata.org/wiki/Q6505497","display_name":"Layer (electronics)","level":2,"score":0.41839998960494995},{"id":"https://openalex.org/C15336307","wikidata":"https://www.wikidata.org/wiki/Q1766051","display_name":"Lens (geology)","level":2,"score":0.41830000281333923},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.39590001106262207},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.38429999351501465},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.31060001254081726},{"id":"https://openalex.org/C192562407","wikidata":"https://www.wikidata.org/wiki/Q228736","display_name":"Materials science","level":0,"score":0.3019999861717224},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.30160000920295715},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.2912999987602234},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.28780001401901245},{"id":"https://openalex.org/C2778751112","wikidata":"https://www.wikidata.org/wiki/Q835016","display_name":"Window (computing)","level":2,"score":0.2635999917984009},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.2597000002861023},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.2547000050544739}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.52202/075280-0521","doi":"https://doi.org/10.52202/075280-0521","is_oa":false,"landing_page_url":"https://doi.org/10.52202/075280-0521","pdf_url":null,"source":{"id":"https://openalex.org/S4363606243","display_name":"neural information processing systems","issn_l":"1049-5258","issn":["1049-5258"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-2"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Advances in Neural Information Processing Systems 36","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2307.11353","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2307.11353","pdf_url":"https://arxiv.org/pdf/2307.11353","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.2307.11353","doi":"https://doi.org/10.48550/arxiv.2307.11353","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2307.11353","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:2307.11353","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2307.11353","pdf_url":"https://arxiv.org/pdf/2307.11353","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":[{"id":"https://openalex.org/G1556123208","display_name":"CIF: SMALL: Theoretical Foundations of Partially Observable Reinforcement Learning: Minimax Sample Complexity and Provably Efficient Algorithms","funder_award_id":"2315725","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G617620878","display_name":"Mean Field Asymptotics in Statistical Inference: Variational Approach, Multiple Testing, and Predictive Inference","funder_award_id":"2210827","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7133201205.pdf","grobid_xml":"https://content.openalex.org/works/W7133201205.grobid-xml"},"referenced_works_count":40,"referenced_works":["https://openalex.org/W2133564696","https://openalex.org/W2586050494","https://openalex.org/W2766371994","https://openalex.org/W2894604724","https://openalex.org/W2896457183","https://openalex.org/W2908479769","https://openalex.org/W2908802752","https://openalex.org/W2950738719","https://openalex.org/W2977969175","https://openalex.org/W2987473824","https://openalex.org/W2994922086","https://openalex.org/W2995273672","https://openalex.org/W3003898522","https://openalex.org/W3035314023","https://openalex.org/W3038074040","https://openalex.org/W3088584794","https://openalex.org/W3091917188","https://openalex.org/W3094502228","https://openalex.org/W3136024375","https://openalex.org/W3165099133","https://openalex.org/W3175990774","https://openalex.org/W3207766955","https://openalex.org/W3208828709","https://openalex.org/W3213602685","https://openalex.org/W4281638513","https://openalex.org/W4297412003","https://openalex.org/W4307868887","https://openalex.org/W4310509152","https://openalex.org/W4311726128","https://openalex.org/W4312107248","https://openalex.org/W4317466024","https://openalex.org/W4318719568","https://openalex.org/W4319991279","https://openalex.org/W4320854821","https://openalex.org/W4321648989","https://openalex.org/W4327810158","https://openalex.org/W4360836968","https://openalex.org/W4377089791","https://openalex.org/W4389775256","https://openalex.org/W7133198840"],"related_works":["https://openalex.org/W4391621801","https://openalex.org/W4411319907","https://openalex.org/W7083807327","https://openalex.org/W7125711612","https://openalex.org/W4414320939","https://openalex.org/W4414897147","https://openalex.org/W4407124429","https://openalex.org/W4414892901","https://openalex.org/W6887891915","https://openalex.org/W4403570399"],"abstract_inverted_index":{"Attention":[0],"layers":[1],"--":[2,14],"which":[3,23],"map":[4],"a":[5,10,35,45,51,57,74,95,101,183,213],"sequence":[6,11,52],"of":[7,12,19,44,53,77,104,175,191,195,202,251],"inputs":[8],"to":[9,111,144],"outputs":[13],"are":[15,108],"core":[16],"building":[17],"blocks":[18],"the":[20,40,65,70,112,145,162,192,196,203,222,242,245,249,252],"Transformer":[21],"architecture":[22],"has":[24,73],"achieved":[25],"significant":[26],"breakthroughs":[27],"in":[28,161,217],"modern":[29],"artificial":[30],"intelligence.":[31],"This":[32],"paper":[33],"presents":[34],"rigorous":[36],"theoretical":[37,237],"study":[38],"on":[39,232],"learning":[41,123,226],"and":[42,56,84,87,172,187,205,239,248],"generalization":[43],"single":[46],"multi-head":[47],"attention":[48,71,97,133,146,185],"layer,":[49],"with":[50,79,134,149,212],"key":[54,85,113,206],"vectors":[55],"separate":[58],"query":[59,83,204],"vector":[60],"as":[61,158],"input.":[62],"We":[63,91,115],"consider":[64],"random":[66,131,151,210],"feature":[67,132,140],"setting":[68],"where":[69,208],"layer":[72,98],"large":[75],"number":[76],"heads,":[78],"randomly":[80],"sampled":[81],"frozen":[82],"matrices,":[86],"trainable":[88],"value":[89],"matrices.":[90],"show":[92],"that":[93,107,177],"such":[94,157],"random-feature":[96,168,184],"can":[99,178],"express":[100],"broad":[102],"class":[103],"target":[105,125,229,253],"functions":[106,126,176],"permutation":[109],"invariant":[110],"vectors.":[114],"further":[116,240],"provide":[117],"quantitative":[118],"excess":[119],"risk":[120],"bounds":[121],"for":[122,154,225],"these":[124],"from":[127],"finite":[128],"samples,":[129],"using":[130],"finitely":[135],"many":[136],"heads.":[137],"Our":[138],"results":[139],"several":[141],"implications":[142],"unique":[143],"structure":[147],"compared":[148],"existing":[150],"features":[152],"theory":[153],"neural":[155],"networks,":[156],"(1)":[159],"Advantages":[160],"sample":[163,219,246],"complexity":[164,250],"over":[165,221],"standard":[166],"two-layer":[167],"networks;":[169],"(2)":[170],"Concrete":[171],"natural":[173,228],"classes":[174],"be":[179],"learned":[180],"efficiently":[181],"by":[182],"layer;":[186],"(3)":[188],"The":[189],"effect":[190],"sampling":[193],"distribution":[194],"query-key":[197],"weight":[198],"matrix":[199],"(the":[200],"product":[201],"matrix),":[207],"Gaussian":[209],"weights":[211],"non-zero":[214],"mean":[215],"result":[216],"better":[218],"complexities":[220],"zero-mean":[223],"counterpart":[224],"certain":[227],"functions.":[230],"Experiments":[231],"simulated":[233],"data":[234],"corroborate":[235],"our":[236],"findings":[238],"illustrate":[241],"interplay":[243],"between":[244],"size":[247],"function.":[254]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":6},{"year":2024,"cited_by_count":6},{"year":2023,"cited_by_count":4}],"updated_date":"2026-11-08T12:12:28.789561","created_date":"2026-03-03T00:00:00"}