{"id":"https://openalex.org/W4391590959","doi":"https://doi.org/10.48550/arxiv.2402.02364","title":"Loss Landscape Degeneracy and Stagewise Development in Transformers","display_name":"Loss Landscape Degeneracy and Stagewise Development in Transformers","publication_year":2024,"publication_date":"2024-02-04","ids":{"openalex":"https://openalex.org/W4391590959","doi":"https://doi.org/10.48550/arxiv.2402.02364"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2402.02364","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.02364","pdf_url":"https://arxiv.org/pdf/2402.02364","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2402.02364","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5104240187","display_name":"Jesse Hoogland","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Hoogland, Jesse","raw_affiliation_strings":["Department of Computer Science , University of Oxford"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , University of Oxford","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Wang, George","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Wang, George","raw_affiliation_strings":["Department of Computer Science , University of Oxford"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , University of Oxford","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5048560932","display_name":"Matthew Farrugia-Roberts","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]},{"id":"https://openalex.org/I56590836","display_name":"Monash University","ror":"https://ror.org/02bfwt286","country_code":"AU","type":"education","lineage":["https://openalex.org/I56590836"]}],"countries":["AU","GB"],"is_corresponding":false,"raw_author_name":"Farrugia-Roberts, Matthew","raw_affiliation_strings":["Department of Computer Science , University of Oxford","Department of Econometrics and Business Statistics , Monash University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , University of Oxford","institution_ids":["https://openalex.org/I40120149"]},{"raw_affiliation_string":"Department of Econometrics and Business Statistics , Monash University","institution_ids":["https://openalex.org/I56590836"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113113922","display_name":"Liam Carroll","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Carroll, Liam","raw_affiliation_strings":["Department of Computer Science , University of Oxford"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , University of Oxford","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5046656905","display_name":"Susan H. Wei","orcid":"https://orcid.org/0000-0002-6842-2352","observed_orcids":["https://orcid.org/0000-0002-6842-2352"]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Wei, Susan","raw_affiliation_strings":["Department of Computer Science , University of Oxford"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , University of Oxford","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5082251796","display_name":"Daniel Murfet","orcid":"https://orcid.org/0000-0003-2495-0070","observed_orcids":["https://orcid.org/0000-0003-2495-0070"]},"institutions":[{"id":"https://openalex.org/I165779595","display_name":"The University of Melbourne","ror":"https://ror.org/01ej9dk98","country_code":"AU","type":"education","lineage":["https://openalex.org/I165779595"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Murfet, Daniel","raw_affiliation_strings":["School of Mathematics and Statistics , the University of Melbourne"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Mathematics and Statistics , the University of Melbourne","institution_ids":["https://openalex.org/I165779595"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"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/T10320","display_name":"Neural Networks and Applications","score":0.7060999870300293,"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/T10320","display_name":"Neural Networks and Applications","score":0.7060999870300293,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.1875,"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":0.013799999840557575,"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/singular-learning-theory","display_name":"singular learning theory","score":0.8889999985694885},{"id":"https://openalex.org/keywords/transformer-based-language-models","display_name":"transformer language models","score":0.6690000295639038},{"id":"https://openalex.org/keywords/model-building","display_name":"model development","score":0.46299999952316284},{"id":"https://openalex.org/keywords/training-dynamics","display_name":"training dynamics","score":0.6039999723434448},{"id":"https://openalex.org/keywords/input-output-behavior","display_name":"input-output behavior","score":0.47699999809265137}],"concepts":[{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5336357355117798},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.42196351289749146},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.41127273440361023},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.3640424609184265},{"id":"https://openalex.org/C166957645","wikidata":"https://www.wikidata.org/wiki/Q23498","display_name":"Archaeology","level":1,"score":0.09138372540473938}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2402.02364","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.02364","pdf_url":"https://arxiv.org/pdf/2402.02364","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2402.02364","doi":"https://doi.org/10.48550/arxiv.2402.02364","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2402.02364","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:2402.02364","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.02364","pdf_url":"https://arxiv.org/pdf/2402.02364","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320320290","display_name":"University of Oxford","ror":"https://ror.org/052gg0110"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4391590959.pdf","grobid_xml":"https://content.openalex.org/works/W4391590959.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4406975771","https://openalex.org/W6927487411","https://openalex.org/W7202147690","https://openalex.org/W4385261255","https://openalex.org/W7214555367","https://openalex.org/W4379474419","https://openalex.org/W4416541469","https://openalex.org/W2959995783","https://openalex.org/W4414990960","https://openalex.org/W3171780980"],"abstract_inverted_index":{"Deep":[0],"learning":[1,44,63,99],"involves":[2],"navigating":[3],"a":[4,37,102,167],"high-dimensional":[5],"loss":[6,81,89,125],"landscape":[7,90,126],"over":[8],"the":[9,15,26,40,49,59,76,80,97,139,144,148,164],"neural":[10,27,52],"network":[11,53],"parameter":[12],"space.":[13],"Over":[14],"course":[16],"of":[17,42,51,61,79,122,147,166],"training,":[18,93],"complex":[19],"computational":[20,141],"structures":[21],"form":[22],"and":[23,55,106,128,143,157],"re-form":[24],"inside":[25],"network,":[28],"leading":[29],"to":[30,45,73],"shifts":[31],"in":[32,75,124,132,138,161],"input/output":[33,145],"behavior.":[34,56],"It":[35],"is":[36,70],"priority":[38],"for":[39,101,170],"science":[41],"deep":[43,173],"uncover":[46],"principles":[47],"governing":[48],"development":[50,69,158],"structure":[54,142],"Drawing":[57],"on":[58],"framework":[60],"singular":[62],"theory,":[64],"we":[65],"propose":[66],"that":[67,114,129,155],"model":[68,105],"deeply":[71],"linked":[72,160],"degeneracy":[74,91,133,156],"local":[77,98],"geometry":[78],"landscape.":[82],"We":[83,112],"investigate":[84],"this":[85],"link":[86],"by":[87,96],"monitoring":[88],"throughout":[92],"as":[94],"quantified":[95],"coefficient,":[100],"transformer":[103],"language":[104],"an":[107],"in-context":[108],"linear":[109],"regression":[110],"transformer.":[111],"show":[113],"training":[115],"can":[116],"be":[117],"divided":[118],"into":[119],"distinct":[120],"periods":[121],"change":[123],"degeneracy,":[127],"these":[130],"changes":[131,137],"coincide":[134],"with":[135],"significant":[136],"internal":[140],"behavior":[146],"transformers.":[149],"This":[150],"finding":[151],"provides":[152],"suggestive":[153],"evidence":[154],"are":[159],"transformers,":[162],"underscoring":[163],"potential":[165],"degeneracy-based":[168],"perspective":[169],"understanding":[171],"modern":[172],"learning.":[174]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-10-14T16:39:01.820549","created_date":"2024-02-07T00:00:00"}