{"id":"https://openalex.org/W7165921304","doi":"https://doi.org/10.48550/arxiv.2606.25010","title":"Emergent Capabilities Arise Randomly from Learning Sparse Attention Patterns","display_name":"Emergent Capabilities Arise Randomly from Learning Sparse Attention Patterns","publication_year":2026,"publication_date":"2026-06-23","ids":{"openalex":"https://openalex.org/W7165921304","doi":"https://doi.org/10.48550/arxiv.2606.25010"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.25010","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.25010","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":"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.25010","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5119915901","display_name":"Vatsal Baherwani","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Baherwani, Vatsal","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139358141","display_name":"Zixi Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Zixi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5061751169","display_name":"Shikai Qiu","orcid":"https://orcid.org/0000-0002-6376-5268"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qiu, Shikai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5107311617","display_name":"A Wilson","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wilson, Andrew Gordon","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5069234607","display_name":"Pavel Izmailov","orcid":"https://orcid.org/0009-0005-0772-2879"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Izmailov, Pavel","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/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.8486999869346619,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.8486999869346619,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"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":0.011500000022351742,"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/T10502","display_name":"Advanced Memory and Neural Computing","score":0.010099999606609344,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.6791999936103821},{"id":"https://openalex.org/keywords/sequence-learning","display_name":"Sequence learning","score":0.4657000005245209},{"id":"https://openalex.org/keywords/feature-learning","display_name":"Feature learning","score":0.462799996137619},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.42340001463890076},{"id":"https://openalex.org/keywords/cellular-automaton","display_name":"Cellular automaton","score":0.3831000030040741},{"id":"https://openalex.org/keywords/context-model","display_name":"Context model","score":0.35339999198913574},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.3255000114440918}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6916000247001648},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6791999936103821},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6061000227928162},{"id":"https://openalex.org/C40506919","wikidata":"https://www.wikidata.org/wiki/Q7452469","display_name":"Sequence learning","level":2,"score":0.4657000005245209},{"id":"https://openalex.org/C59404180","wikidata":"https://www.wikidata.org/wiki/Q17013334","display_name":"Feature learning","level":2,"score":0.462799996137619},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.42340001463890076},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.38580000400543213},{"id":"https://openalex.org/C35527583","wikidata":"https://www.wikidata.org/wiki/Q189156","display_name":"Cellular automaton","level":2,"score":0.3831000030040741},{"id":"https://openalex.org/C183322885","wikidata":"https://www.wikidata.org/wiki/Q17007702","display_name":"Context model","level":3,"score":0.35339999198913574},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.3255000114440918},{"id":"https://openalex.org/C160920958","wikidata":"https://www.wikidata.org/wiki/Q7662746","display_name":"Synthetic data","level":2,"score":0.31610000133514404},{"id":"https://openalex.org/C33676613","wikidata":"https://www.wikidata.org/wiki/Q13415176","display_name":"Dimension (graph theory)","level":2,"score":0.2928999960422516},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.28450000286102295},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.27570000290870667},{"id":"https://openalex.org/C64876066","wikidata":"https://www.wikidata.org/wiki/Q5141226","display_name":"Cognitive neuroscience of visual object recognition","level":3,"score":0.26739999651908875},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2556999921798706},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.25540000200271606}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.25010","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.25010","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.25010","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.25010","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":"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"},"sustainable_development_goals":[{"score":0.5572512149810791,"id":"https://metadata.un.org/sdg/4","display_name":"Quality Education"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Neural":[0],"scaling":[1,111],"laws":[2],"for":[3],"transformer":[4,83,149,183],"language":[5],"models":[6,47,84],"predict":[7],"smooth":[8],"improvements":[9],"in":[10,182],"pretraining":[11],"loss":[12],"with":[13,45,140,154],"increasing":[14,125],"parameters,":[15],"but":[16],"downstream":[17,168],"capabilities":[18,40,59,169],"such":[19,60],"as":[20,61],"in-context":[21],"learning":[22,72,100,118,178],"are":[23],"known":[24],"to":[25,69,173],"emerge":[26],"abruptly":[27,171],"past":[28,132],"a":[29,133,148,161],"certain":[30],"model":[31],"scale.":[32],"In":[33],"this":[34,79],"paper,":[35],"we":[36,81,94],"show":[37,95],"that":[38,55,96,145,167],"emergent":[39],"arise":[41,170],"stochastically":[42],"throughout":[43],"training,":[44],"larger":[46],"acquiring":[48],"them":[49],"earlier":[50],"on":[51,85,104,120,150],"average.":[52],"We":[53,136],"demonstrate":[54],"the":[56,70,97,112,126,174],"emergence":[57],"of":[58,73,99,114,177],"pattern":[62,108],"completion":[63],"and":[64,89,93,107],"indirect":[65],"object":[66],"identification":[67],"corresponds":[68],"abrupt":[71],"task-relevant":[74],"attention":[75,101,115,142,156,180],"patterns.":[76,157],"To":[77],"isolate":[78],"phenomenon,":[80],"train":[82],"synthetic":[86,122],"linear":[87,151],"map":[88,152],"cellular":[90],"automata":[91],"datasets,":[92],"difficulty":[98,176],"patterns":[102,181],"depends":[103],"context":[105],"length":[106],"sparsity.":[109],"Moreover,":[110],"number":[113],"heads":[116],"improves":[117],"efficiency":[119],"our":[121],"tasks,":[123],"while":[124],"head":[127],"dimension":[128],"yields":[129],"diminishing":[130],"returns":[131],"minimum":[134],"capacity.":[135],"additionally":[137],"investigate":[138],"architectures":[139],"alternative":[141],"mechanisms,":[143],"showing":[144,166],"MLP-Mixer":[146],"outperforms":[147],"tasks":[153],"complex":[155],"Our":[158],"findings":[159],"provide":[160],"mechanistic":[162],"insight":[163],"into":[164],"emergence,":[165],"due":[172],"intrinsic":[175],"sparse":[179],"models.":[184]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-26T00:00:00"}