{"id":"https://openalex.org/W4366731970","doi":"https://doi.org/10.48550/arxiv.2304.09871","title":"A Theory on Adam Instability in Large-Scale Machine Learning","display_name":"A Theory on Adam Instability in Large-Scale Machine Learning","publication_year":2023,"publication_date":"2023-04-19","ids":{"openalex":"https://openalex.org/W4366731970","doi":"https://doi.org/10.48550/arxiv.2304.09871"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2304.09871","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2304.09871","pdf_url":"https://arxiv.org/pdf/2304.09871","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/2304.09871","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5041560478","display_name":"Igor Molybog","orcid":"https://orcid.org/0000-0001-7251-3532","observed_orcids":["https://orcid.org/0000-0001-7251-3532"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Molybog, Igor","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Albert, Peter","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Albert, Peter","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5057089836","display_name":"Moya Chen","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Moya","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5008830900","display_name":"Zachary DeVito","orcid":"https://orcid.org/0009-0002-8863-1503","observed_orcids":["https://orcid.org/0009-0002-8863-1503"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"DeVito, Zachary","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5060413371","display_name":"David Esiobu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Esiobu, David","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5075834790","display_name":"Naman Goyal","orcid":"https://orcid.org/0000-0002-7565-4303","observed_orcids":["https://orcid.org/0000-0002-7565-4303"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Goyal, Naman","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5051808640","display_name":"Punit Singh Koura","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Koura, Punit Singh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5079540764","display_name":"Sharan Narang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Narang, Sharan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Poulton, Andrew","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Poulton, Andrew","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5000326689","display_name":"Ruan Silva","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Silva, Ruan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5033021890","display_name":"Binh Tang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tang, Binh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5005549348","display_name":"Diana Liskovich","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liskovich, Diana","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5048354069","display_name":"Puxin Xu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Puxin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100424317","display_name":"Yuchen Zhang","orcid":"https://orcid.org/0000-0001-8536-7175","observed_orcids":["https://orcid.org/0000-0001-8536-7175","https://orcid.org/0009-0008-9178-7369"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Yuchen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5024510591","display_name":"Melanie Kambadur","orcid":"https://orcid.org/0009-0008-7481-6850","observed_orcids":["https://orcid.org/0009-0008-7481-6850"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kambadur, Melanie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5108644922","display_name":"Stephen Roller","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Roller, Stephen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5063484880","display_name":"Susan Zhang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Susan","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":true,"cited_by_count":2,"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":0.9945999979972839,"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":0.9945999979972839,"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/T10028","display_name":"Topic Modeling","score":0.0019000000320374966,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.0003000000142492354,"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/adam-optimizer","display_name":"Adam optimizer","score":0.8679999709129333},{"id":"https://openalex.org/keywords/training-instability","display_name":"training instability","score":0.4580000042915344},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.703000009059906},{"id":"https://openalex.org/keywords/loss-landscape","display_name":"loss landscape","score":0.6990000009536743},{"id":"https://openalex.org/keywords/deep-learning-training","display_name":"deep learning training","score":0.5509999990463257}],"concepts":[{"id":"https://openalex.org/C2779010991","wikidata":"https://www.wikidata.org/wiki/Q2720909","display_name":"Artifact (error)","level":2,"score":0.5622739195823669},{"id":"https://openalex.org/C207390915","wikidata":"https://www.wikidata.org/wiki/Q1230525","display_name":"Divergence (linguistics)","level":2,"score":0.5381321310997009},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.4908502697944641},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.47768956422805786},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.47201889753341675},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.45909106731414795},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.45782431960105896},{"id":"https://openalex.org/C169345407","wikidata":"https://www.wikidata.org/wiki/Q8216221","display_name":"Uncorrelated","level":2,"score":0.4418778419494629},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3794284462928772},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.3230111002922058},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2292294204235077},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.1978735327720642},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.17143648862838745},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.15149185061454773},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.1413687765598297},{"id":"https://openalex.org/C58640448","wikidata":"https://www.wikidata.org/wiki/Q42515","display_name":"Cartography","level":1,"score":0.1022132933139801},{"id":"https://openalex.org/C153294291","wikidata":"https://www.wikidata.org/wiki/Q25261","display_name":"Meteorology","level":1,"score":0.08080059289932251}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2304.09871","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2304.09871","pdf_url":"https://arxiv.org/pdf/2304.09871","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.2304.09871","doi":"https://doi.org/10.48550/arxiv.2304.09871","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2304.09871","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:2304.09871","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2304.09871","pdf_url":"https://arxiv.org/pdf/2304.09871","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":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4366731970.pdf","grobid_xml":"https://content.openalex.org/works/W4366731970.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7133309390","https://openalex.org/W7057822595","https://openalex.org/W7160931171","https://openalex.org/W4415111888","https://openalex.org/W7126081955","https://openalex.org/W4405767421","https://openalex.org/W7127359426","https://openalex.org/W7212715725","https://openalex.org/W7084586124","https://openalex.org/W7165612465"],"abstract_inverted_index":{"We":[0,18,36],"present":[1,107],"a":[2,42,51,84,88],"theory":[3],"for":[4,32],"the":[5,12,21,27,46,60,65,81,94,104,110,114],"previously":[6],"unexplained":[7],"divergent":[8],"behavior":[9],"noticed":[10],"in":[11,44,80],"training":[13,66,82,111],"of":[14,26,62,83,97,113,117],"large":[15,53,89],"language":[16,99,115],"models.":[17],"argue":[19,103],"that":[20,38],"phenomenon":[22],"is":[23,56,74,93],"an":[24],"artifact":[25,73],"dominant":[28],"optimization":[29],"algorithm":[30],"used":[31],"training,":[33],"called":[34],"Adam.":[35],"observe":[37],"Adam":[39],"can":[40],"enter":[41],"state":[43],"which":[45,92],"parameter":[47],"update":[48],"vector":[49],"has":[50],"relatively":[52],"norm":[54],"and":[55,126],"essentially":[57],"uncorrelated":[58],"with":[59,87],"direction":[61],"descent":[63],"on":[64],"loss":[67],"landscape,":[68],"leading":[69],"to":[70,77],"divergence.":[71],"This":[72],"more":[75],"likely":[76],"be":[78],"observed":[79],"deep":[85],"model":[86,100],"batch":[90],"size,":[91],"typical":[95],"setting":[96],"large-scale":[98],"training.":[101],"To":[102],"theory,":[105],"we":[106],"observations":[108],"from":[109],"runs":[112],"models":[116],"different":[118],"scales:":[119],"7":[120],"billion,":[121,123,125],"30":[122],"65":[124],"546":[127],"billion":[128],"parameters.":[129]},"counts_by_year":[{"year":2023,"cited_by_count":2}],"updated_date":"2026-10-16T09:59:22.235173","created_date":"2023-04-24T00:00:00"}