{"id":"https://openalex.org/W4318719486","doi":"https://doi.org/10.48550/arxiv.2301.13104","title":"Equivariant Differentially Private Deep Learning: Why DP-SGD Needs Sparser Models","display_name":"Equivariant Differentially Private Deep Learning: Why DP-SGD Needs Sparser Models","publication_year":2023,"publication_date":"2023-01-30","ids":{"openalex":"https://openalex.org/W4318719486","doi":"https://doi.org/10.48550/arxiv.2301.13104"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2301.13104","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2301.13104","pdf_url":"https://arxiv.org/pdf/2301.13104","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"","raw_type":"text"},"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2301.13104","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5072127418","display_name":"Florian A. H\u00f6lzl","orcid":"https://orcid.org/0000-0003-3218-7976"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"H\u00f6lzl, Florian A.","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5006461848","display_name":"Daniel Rueckert","orcid":"https://orcid.org/0000-0002-5683-5889"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rueckert, Daniel","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5000554104","display_name":"Georgios Kaissis","orcid":"https://orcid.org/0000-0001-8382-8062"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kaissis, Georgios","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":3,"corresponding_author_ids":["https://openalex.org/A5072127418"],"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/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.9998999834060669,"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/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.9998999834060669,"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":0.9947999715805054,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.973800003528595,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/computer-science","display_name":"Computer science","score":0.7796354293823242},{"id":"https://openalex.org/keywords/stochastic-gradient-descent","display_name":"Stochastic gradient descent","score":0.6272045373916626},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.6095248460769653},{"id":"https://openalex.org/keywords/differential-privacy","display_name":"Differential privacy","score":0.5660014152526855},{"id":"https://openalex.org/keywords/deep-learning","display_name":"Deep learning","score":0.5648815631866455},{"id":"https://openalex.org/keywords/machine-learning","display_name":"Machine learning","score":0.5400133728981018},{"id":"https://openalex.org/keywords/rendering","display_name":"Rendering (computer graphics)","score":0.472623348236084},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.44304215908050537},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.4414079785346985},{"id":"https://openalex.org/keywords/computer-engineering","display_name":"Computer engineering","score":0.42026960849761963},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.24085742235183716},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.2271701693534851}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7796354293823242},{"id":"https://openalex.org/C206688291","wikidata":"https://www.wikidata.org/wiki/Q7617819","display_name":"Stochastic gradient descent","level":3,"score":0.6272045373916626},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6095248460769653},{"id":"https://openalex.org/C23130292","wikidata":"https://www.wikidata.org/wiki/Q5275358","display_name":"Differential privacy","level":2,"score":0.5660014152526855},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.5648815631866455},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.5400133728981018},{"id":"https://openalex.org/C205711294","wikidata":"https://www.wikidata.org/wiki/Q176953","display_name":"Rendering (computer graphics)","level":2,"score":0.472623348236084},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.44304215908050537},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.4414079785346985},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.42026960849761963},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.24085742235183716},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.2271701693534851},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2301.13104","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2301.13104","pdf_url":"https://arxiv.org/pdf/2301.13104","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"","raw_type":"text"},{"id":"doi:10.48550/arxiv.2301.13104","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2301.13104","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":"article"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2301.13104","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2301.13104","pdf_url":"https://arxiv.org/pdf/2301.13104","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"","raw_type":"text"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/9","score":0.47999998927116394,"display_name":"Industry, innovation and infrastructure"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4318719486.pdf"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W3038283795","https://openalex.org/W2604501336","https://openalex.org/W2734500670","https://openalex.org/W2558166297","https://openalex.org/W2315671126","https://openalex.org/W798507144","https://openalex.org/W2964481303","https://openalex.org/W1751413323","https://openalex.org/W2571704763","https://openalex.org/W4361791424"],"abstract_inverted_index":{"Differentially":[0],"Private":[1],"Stochastic":[2],"Gradient":[3],"Descent":[4],"(DP-SGD)":[5],"limits":[6],"the":[7,28,87,103,184,209],"amount":[8],"of":[9,90,105,176,204],"private":[10,213],"information":[11],"deep":[12,216],"learning":[13,48,217],"models":[14,45,164],"can":[15,160],"memorize":[16],"during":[17],"training.":[18],"This":[19],"is":[20],"achieved":[21],"by":[22,136,138,187],"clipping":[23],"and":[24,31,78,92,120,156,207,214],"adding":[25],"noise":[26],"to":[27,109,178],"model's":[29],"gradients,":[30],"thus":[32,124],"networks":[33,142],"with":[34,53,115,146,165],"more":[35,58,188],"parameters":[36,206],"require":[37],"proportionally":[38],"stronger":[39],"perturbation.":[40],"As":[41],"a":[42,127,194],"result,":[43],"large":[44,79],"have":[46],"difficulties":[47],"useful":[49],"information,":[50],"rendering":[51],"training":[52,60,71,145],"DP-SGD":[54,91],"exceedingly":[55],"difficult":[56],"on":[57,66],"challenging":[59],"tasks.":[61],"Recent":[62],"research":[63],"has":[64],"focused":[65],"combating":[67],"this":[68,98],"challenge":[69],"through":[70],"adaptations":[72],"such":[73,112,134],"as":[74,126],"heavy":[75],"data":[76],"augmentation":[77],"batch":[80],"sizes.":[81],"However,":[82],"these":[83],"techniques":[84],"further":[85],"increase":[86,175],"computational":[88,168],"overhead":[89],"reduce":[93],"its":[94],"practical":[95],"applicability.":[96],"In":[97],"work,":[99],"we":[100,152,172],"propose":[101],"using":[102],"principle":[104],"sparse":[106],"model":[107,144,198],"design":[108,137,159],"solve":[110],"precisely":[111],"complex":[113],"tasks":[114],"fewer":[116],"parameters,":[117],"higher":[118],"accuracy,":[119],"in":[121,180],"less":[122],"time,":[123],"serving":[125],"promising":[128],"direction":[129],"for":[130,143,218],"DP-SGD.":[131],"We":[132],"achieve":[133,173],"sparsity":[135],"introducing":[139],"equivariant":[140,150],"convolutional":[141],"Differential":[147],"Privacy.":[148],"Using":[149],"networks,":[151],"show":[153],"that":[154,200],"small":[155],"efficient":[157,197],"architecture":[158],"outperform":[161],"current":[162],"state-of-the-art":[163],"substantially":[166],"lower":[167],"requirements.":[169],"On":[170],"CIFAR-10,":[171],"an":[174],"up":[177],"$9\\%$":[179],"accuracy":[181],"while":[182],"reducing":[183],"computation":[185],"time":[186],"than":[189],"$85\\%$.":[190],"Our":[191],"results":[192],"are":[193],"step":[195],"towards":[196],"architectures":[199],"make":[201],"optimal":[202],"use":[203],"their":[205],"bridge":[208],"privacy-utility":[210],"gap":[211],"between":[212],"non-private":[215],"computer":[219],"vision.":[220]},"counts_by_year":[{"year":2024,"cited_by_count":1}],"updated_date":"2026-03-11T14:59:36.786465","created_date":"2023-02-01T00:00:00"}