{"id":"https://openalex.org/W2785648721","doi":"https://doi.org/10.48550/arxiv.1803.06959","title":"On the importance of single directions for generalization","display_name":"On the importance of single directions for generalization","publication_year":2018,"publication_date":"2018-03-19","ids":{"openalex":"https://openalex.org/W2785648721","doi":"https://doi.org/10.48550/arxiv.1803.06959","mag":"2785648721"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:1803.06959","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/1803.06959","pdf_url":"https://arxiv.org/pdf/1803.06959","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"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/1803.06959","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5042062822","display_name":"Ari S. Morcos","orcid":"https://orcid.org/0000-0001-5370-6515","observed_orcids":["https://orcid.org/0000-0001-5370-6515"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Morcos, Ari S.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5109502776","display_name":"David G. T. Barrett","orcid":"https://orcid.org/0009-0004-4678-6691","observed_orcids":["https://orcid.org/0009-0004-4678-6691"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Barrett, David G. T.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010900024","display_name":"Neil C. Rabinowitz","orcid":"https://orcid.org/0000-0002-4030-1567","observed_orcids":["https://orcid.org/0000-0002-4030-1567"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rabinowitz, Neil C.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5083771180","display_name":"Matthew Botvinick","orcid":"https://orcid.org/0000-0001-7758-6896","observed_orcids":["https://orcid.org/0000-0001-7758-6896","https://orcid.org/0009-0000-1902-4000"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Botvinick, Matthew","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":306,"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":0.44510000944137573,"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":0.44510000944137573,"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":0.2393999993801117,"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":0.0949999988079071,"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/generalization","display_name":"generalization","score":0.8489999771118164},{"id":"https://openalex.org/keywords/deep-neural-network","display_name":"deep neural network","score":0.7429999709129333},{"id":"https://openalex.org/keywords/noisy-labels","display_name":"label noise","score":0.6140000224113464},{"id":"https://openalex.org/keywords/ensembles","display_name":"ensembles","score":0.47699999809265137},{"id":"https://openalex.org/keywords/dropout","display_name":"dropout","score":0.7509999871253967},{"id":"https://openalex.org/keywords/batch-normalization","display_name":"batch normalization","score":0.9990000128746033},{"id":"https://openalex.org/keywords/task-significance","display_name":"task significance","score":0.5759999752044678},{"id":"https://openalex.org/keywords/training-dynamics","display_name":"training dynamics","score":0.3240000009536743}],"concepts":[{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.7907189130783081},{"id":"https://openalex.org/C144237770","wikidata":"https://www.wikidata.org/wiki/Q747534","display_name":"Mathematical economics","level":1,"score":0.4122655391693115},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.33548054099082947},{"id":"https://openalex.org/C111472728","wikidata":"https://www.wikidata.org/wiki/Q9471","display_name":"Epistemology","level":1,"score":0.2252711057662964},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.17474672198295593}],"mesh":[],"locations_count":3,"locations":[{"id":"pmh:oai:arXiv.org:1803.06959","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/1803.06959","pdf_url":"https://arxiv.org/pdf/1803.06959","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.1803.06959","doi":"https://doi.org/10.48550/arxiv.1803.06959","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.1803.06959","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"},{"id":"mag:2963420658","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/pdf/1803.06959","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":"arXiv (Cornell University)","raw_type":null}],"best_oa_location":{"id":"pmh:oai:arXiv.org:1803.06959","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/1803.06959","pdf_url":"https://arxiv.org/pdf/1803.06959","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":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W2785648721.pdf","grobid_xml":"https://content.openalex.org/works/W2785648721.grobid-xml"},"referenced_works_count":31,"referenced_works":["https://openalex.org/W1849277567","https://openalex.org/W1899185266","https://openalex.org/W1975145078","https://openalex.org/W1998050452","https://openalex.org/W2027914108","https://openalex.org/W2033708181","https://openalex.org/W2039611165","https://openalex.org/W2047125104","https://openalex.org/W2111096131","https://openalex.org/W2150367061","https://openalex.org/W2160921898","https://openalex.org/W2194775991","https://openalex.org/W2433379750","https://openalex.org/W2518631572","https://openalex.org/W2525778437","https://openalex.org/W2527978657","https://openalex.org/W2529194139","https://openalex.org/W2560805033","https://openalex.org/W2593634001","https://openalex.org/W2604117713","https://openalex.org/W2606347107","https://openalex.org/W2610018085","https://openalex.org/W2622255292","https://openalex.org/W2682189153","https://openalex.org/W2765733029","https://openalex.org/W2767204723","https://openalex.org/W2809994596","https://openalex.org/W2949117887","https://openalex.org/W2963208657","https://openalex.org/W2963739978","https://openalex.org/W3100682276"],"related_works":["https://openalex.org/W2903118043","https://openalex.org/W6960063051","https://openalex.org/W3131466122","https://openalex.org/W3008067209","https://openalex.org/W2937266760","https://openalex.org/W4383173863","https://openalex.org/W4381786116","https://openalex.org/W2944572902","https://openalex.org/W2295524297","https://openalex.org/W4388890003"],"abstract_inverted_index":{"Despite":[0],"their":[1,63,172,179],"ability":[2],"to":[3,56,76,130],"memorize":[4],"large":[5],"datasets,":[6],"deep":[7],"neural":[8],"networks":[9,23,94,107,167],"often":[10],"achieve":[11],"good":[12,87],"generalization":[13,91],"performance.":[14,194],"However,":[15],"the":[16,19,34,42,119,144],"differences":[17],"between":[18],"learned":[20],"solutions":[21],"of":[22,37,44,52,74,89,101,106,121,147,160],"which":[24,28,168],"generalize":[25,169],"and":[26,117],"those":[27],"do":[29],"not":[30,66,164,188],"remain":[31],"unclear.":[32],"Additionally,":[33],"tuning":[35],"properties":[36],"single":[38,46,83,137],"directions":[39,84],"(defined":[40],"as":[41],"activation":[43],"a":[45,79,86,131,157],"unit":[47],"or":[48],"some":[49,57],"linear":[50],"combination":[51],"units":[53,176,186],"in":[54,140],"response":[55],"input)":[58],"have":[59],"been":[60,67],"highlighted,":[61],"but":[62,181],"importance":[64],"has":[65],"evaluated.":[68],"Here,":[69],"we":[70,151],"connect":[71],"these":[72],"lines":[73],"inquiry":[75],"demonstrate":[77],"that":[78,153,166,183],"network's":[80],"reliance":[81],"on":[82,96,109,174],"is":[85,156],"predictor":[88,159],"its":[90],"performance,":[92],"across":[93,104,114],"trained":[95,108],"datasets":[97,110],"with":[98,111],"different":[99,115],"fractions":[100],"corrupted":[102],"labels,":[103,113],"ensembles":[105],"unmodified":[112],"hyperparameters,":[116],"over":[118],"course":[120],"training.":[122],"While":[123],"dropout":[124],"only":[125,165],"regularizes":[126],"this":[127],"quantity":[128],"up":[129],"point,":[132],"batch":[133],"normalization":[134],"implicitly":[135],"discourages":[136],"direction":[138],"reliance,":[139],"part":[141],"by":[142,177],"decreasing":[143],"class":[145,154],"selectivity":[146,155],"individual":[148,175],"units.":[149],"Finally,":[150],"find":[152],"poor":[158],"task":[161],"importance,":[162],"suggesting":[163],"well":[170],"minimize":[171],"dependence":[173],"reducing":[178],"selectivity,":[180],"also":[182],"individually":[184],"selective":[185],"may":[187],"be":[189],"necessary":[190],"for":[191],"strong":[192],"network":[193]},"counts_by_year":[{"year":2026,"cited_by_count":7},{"year":2025,"cited_by_count":18},{"year":2024,"cited_by_count":19},{"year":2023,"cited_by_count":31},{"year":2022,"cited_by_count":29},{"year":2021,"cited_by_count":59},{"year":2020,"cited_by_count":62},{"year":2019,"cited_by_count":49},{"year":2018,"cited_by_count":31}],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2018-02-23T00:00:00"}