{"id":"https://openalex.org/W7166687380","doi":"https://doi.org/10.48550/arxiv.2606.30444","title":"SGD Provably Prioritizes a Shortcut Spurious Feature in the XOR Model","display_name":"SGD Provably Prioritizes a Shortcut Spurious Feature in the XOR Model","publication_year":2026,"publication_date":"2026-06-29","ids":{"openalex":"https://openalex.org/W7166687380","doi":"https://doi.org/10.48550/arxiv.2606.30444"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2606.30444","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2606.30444","pdf_url":"https://arxiv.org/pdf/2606.30444","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/2606.30444","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5036834037","display_name":"Tyler LaBonte","orcid":"https://orcid.org/0000-0002-3781-7212","observed_orcids":["https://orcid.org/0000-0002-3781-7212"]},"institutions":[{"id":"https://openalex.org/I130701444","display_name":"Georgia Institute of Technology","ror":"https://ror.org/01zkghx44","country_code":"US","type":"education","lineage":["https://openalex.org/I130701444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"LaBonte, Tyler","raw_affiliation_strings":["School of Industrial and Systems Engineering , Georgia Institute of Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Industrial and Systems Engineering , Georgia Institute of Technology","institution_ids":["https://openalex.org/I130701444"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5103255732","display_name":"Vidya Muthukumar","orcid":"https://orcid.org/0000-0003-2786-7360","observed_orcids":["https://orcid.org/0000-0003-2786-7360"]},"institutions":[{"id":"https://openalex.org/I130701444","display_name":"Georgia Institute of Technology","ror":"https://ror.org/01zkghx44","country_code":"US","type":"education","lineage":["https://openalex.org/I130701444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Muthukumar, Vidya","raw_affiliation_strings":["School of Electrical and Computer Engineering , Georgia Institute of Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Electrical and Computer Engineering , Georgia Institute of Technology","institution_ids":["https://openalex.org/I130701444"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I130701444"],"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.98580002784729,"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":0.98580002784729,"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":0.009499999694526196,"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":0.0008999999845400453,"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/spurious-correlations","display_name":"spurious correlations","score":0.8500000238418579},{"id":"https://openalex.org/keywords/shortcut-learning","display_name":"shortcut learning","score":0.871999979019165},{"id":"https://openalex.org/keywords/xor-problem","display_name":"XOR problem","score":0.5680000185966492},{"id":"https://openalex.org/keywords/optimization-dynamics","display_name":"optimization dynamics","score":0.6800000071525574},{"id":"https://openalex.org/keywords/boolean-hypercube","display_name":"Boolean hypercube","score":0.7129999995231628},{"id":"https://openalex.org/keywords/phase-transitions","display_name":"phase transitions","score":0.6380000114440918},{"id":"https://openalex.org/keywords/sample-complexity","display_name":"sample complexity","score":0.6850000023841858}],"concepts":[{"id":"https://openalex.org/C97256817","wikidata":"https://www.wikidata.org/wiki/Q1462316","display_name":"Spurious relationship","level":2,"score":0.9818000197410583},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.7739999890327454},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5773000121116638},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5353999733924866},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.4948999881744385},{"id":"https://openalex.org/C774472","wikidata":"https://www.wikidata.org/wiki/Q6760393","display_name":"Margin (machine learning)","level":2,"score":0.4803999960422516},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4366999864578247},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.4343999922275543},{"id":"https://openalex.org/C2779843651","wikidata":"https://www.wikidata.org/wiki/Q7390335","display_name":"SIGNAL (programming language)","level":2,"score":0.3984000086784363},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.3517000079154968},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3513000011444092},{"id":"https://openalex.org/C52622490","wikidata":"https://www.wikidata.org/wiki/Q1026626","display_name":"Feature extraction","level":2,"score":0.33500000834465027},{"id":"https://openalex.org/C59404180","wikidata":"https://www.wikidata.org/wiki/Q17013334","display_name":"Feature learning","level":2,"score":0.31200000643730164},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2994999885559082},{"id":"https://openalex.org/C104267543","wikidata":"https://www.wikidata.org/wiki/Q208163","display_name":"Signal processing","level":3,"score":0.29409998655319214},{"id":"https://openalex.org/C117220453","wikidata":"https://www.wikidata.org/wiki/Q5172842","display_name":"Correlation","level":2,"score":0.2799000144004822},{"id":"https://openalex.org/C187455244","wikidata":"https://www.wikidata.org/wiki/Q942353","display_name":"Boolean function","level":2,"score":0.2612999975681305},{"id":"https://openalex.org/C198531522","wikidata":"https://www.wikidata.org/wiki/Q485146","display_name":"Sample (material)","level":2,"score":0.2556000053882599}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2606.30444","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2606.30444","pdf_url":"https://arxiv.org/pdf/2606.30444","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.2606.30444","doi":"https://doi.org/10.48550/arxiv.2606.30444","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.30444","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:2606.30444","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2606.30444","pdf_url":"https://arxiv.org/pdf/2606.30444","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/G3442346314","display_name":"CIF: RI: Medium: Design principles and theory for data augmentation","funder_award_id":"2212182","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G7821036309","display_name":"CAREER: Overparameterization in modern machine learning: A panacea or a pitfall?","funder_award_id":"2239151","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/W7166687380.pdf","grobid_xml":"https://content.openalex.org/works/W7166687380.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4392575412","https://openalex.org/W4378942776","https://openalex.org/W4387156670","https://openalex.org/W7087715124","https://openalex.org/W4377866414","https://openalex.org/W4226426379","https://openalex.org/W4391709409","https://openalex.org/W7220958358","https://openalex.org/W4415873422","https://openalex.org/W7133191711"],"abstract_inverted_index":{"Neural":[0],"networks":[1,60],"are":[2],"known":[3],"to":[4,7,28],"be":[5,194],"susceptible":[6],"over-reliance":[8],"on":[9,33,66],"spurious":[10,53,89,97,109,116,142,152,172,182,201,229],"correlations.":[11],"However,":[12],"the":[13,39,47,67,75,96,104,108,129,133,138,141,151,155,165,171,181,187,200,208,219,224,228],"precise":[14,125],"mechanism":[15],"by":[16,62],"which":[17],"models":[18],"exploit":[19],"shortcut":[20],"features":[21],"is":[22,174,211,231],"not":[23,232],"fully":[24],"understood,":[25],"and":[26,86,100,110,144,164],"algorithms":[27],"mitigate":[29],"this":[30,43],"behavior":[31],"rely":[32],"as":[34],"yet":[35],"unjustified":[36],"assumptions":[37],"about":[38],"learned":[40,195],"representations.":[41],"In":[42,132,154,205],"work,":[44],"we":[45,177,213],"provide":[46,214],"first":[48,134],"end-to-end":[49],"theoretical":[50],"characterization":[51],"of":[52,140,150],"feature":[54,98,120,143,167,183,202,230],"learning":[55,130,163],"for":[56],"two-layer":[57],"ReLU":[58],"neural":[59],"trained":[61],"online":[63],"minibatch":[64],"SGD":[65,94],"logistic":[68],"loss.":[69],"We":[70,91],"consider":[71],"data":[72],"drawn":[73],"from":[74],"high-dimensional":[76],"Boolean":[77],"hypercube":[78],"with":[79,113],"a":[80,87,114],"quadratic":[81],"signal":[82,111,119,166],"function":[83],"(namely":[84],"XOR)":[85],"linear":[88],"correlation.":[90],"show":[92,178],"that":[93,180,218],"learns":[95],"first,":[99],"exponentially":[101],"fast.":[102],"Moreover,":[103],"optimization":[105],"dynamics":[106],"couple":[107],"features,":[112],"stronger":[115],"component":[117],"inhibiting":[118],"learning.":[121],"Our":[122],"analysis":[123],"reveals":[124],"phase":[126],"transitions":[127],"in":[128,196],"dynamics.":[131],"phase,":[135,157],"alignment":[136],"between":[137],"signs":[139],"second-layer":[145],"weight":[146],"drives":[147],"rapid":[148],"growth":[149],"feature.":[153],"second":[156],"large":[158],"majority":[159],"group":[160],"margin":[161],"slows":[162],"remains":[168],"suppressed.":[169],"When":[170],"correlation":[173,209],"maximally":[175],"strong,":[176],"theoretically":[179],"dominates":[184],"even":[185],"at":[186],"sample":[188],"complexity":[189],"threshold":[190],"where":[191],"XOR":[192,225],"would":[193],"isolation":[197],"(i.e.,":[198],"if":[199],"was":[203],"absent).":[204],"contrast,":[206],"when":[207],"strength":[210],"constant,":[212],"preliminary":[215],"empirical":[216],"evidence":[217],"model":[220],"can":[221],"eventually":[222],"learn":[223],"signal,":[226],"although":[227],"forgotten.":[233]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-07-01T00:00:00"}