{"id":"https://openalex.org/W7160835706","doi":"https://doi.org/10.48550/arxiv.2605.07072","title":"Less Random, More Private: What is the Optimal Subsampling Scheme for DP-SGD?","display_name":"Less Random, More Private: What is the Optimal Subsampling Scheme for DP-SGD?","publication_year":2026,"publication_date":"2026-05-08","ids":{"openalex":"https://openalex.org/W7160835706","doi":"https://doi.org/10.48550/arxiv.2605.07072"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.07072","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.07072","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,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.07072","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5059957550","display_name":"Andy Dong","orcid":"https://orcid.org/0000-0001-6975-4466"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dong, Andy","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135900894","display_name":"Ayfer \u00d6zg\u00fcr","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"\u00d6zg\u00fcr, Ayfer","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":false,"cited_by_count":0,"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":0.9681000113487244,"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":0.9681000113487244,"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":0.013500000350177288,"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/T10237","display_name":"Cryptography and Data Security","score":0.00559999980032444,"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/randomness","display_name":"Randomness","score":0.8377000093460083},{"id":"https://openalex.org/keywords/poisson-distribution","display_name":"Poisson distribution","score":0.6043999791145325},{"id":"https://openalex.org/keywords/variance","display_name":"Variance (accounting)","score":0.5062000155448914},{"id":"https://openalex.org/keywords/monte-carlo-method","display_name":"Monte Carlo method","score":0.491100013256073},{"id":"https://openalex.org/keywords/multiplier","display_name":"Multiplier (economics)","score":0.48570001125335693},{"id":"https://openalex.org/keywords/estimator","display_name":"Estimator","score":0.44690001010894775},{"id":"https://openalex.org/keywords/intuition","display_name":"Intuition","score":0.4415999948978424},{"id":"https://openalex.org/keywords/sampling-scheme","display_name":"Sampling scheme","score":0.4156999886035919},{"id":"https://openalex.org/keywords/noise","display_name":"Noise (video)","score":0.41280001401901245}],"concepts":[{"id":"https://openalex.org/C125112378","wikidata":"https://www.wikidata.org/wiki/Q176640","display_name":"Randomness","level":2,"score":0.8377000093460083},{"id":"https://openalex.org/C100906024","wikidata":"https://www.wikidata.org/wiki/Q205692","display_name":"Poisson distribution","level":2,"score":0.6043999791145325},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5149999856948853},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.5062000155448914},{"id":"https://openalex.org/C19499675","wikidata":"https://www.wikidata.org/wiki/Q232207","display_name":"Monte Carlo method","level":2,"score":0.491100013256073},{"id":"https://openalex.org/C124584101","wikidata":"https://www.wikidata.org/wiki/Q1053266","display_name":"Multiplier (economics)","level":2,"score":0.48570001125335693},{"id":"https://openalex.org/C185429906","wikidata":"https://www.wikidata.org/wiki/Q1130160","display_name":"Estimator","level":2,"score":0.44690001010894775},{"id":"https://openalex.org/C132010649","wikidata":"https://www.wikidata.org/wiki/Q189222","display_name":"Intuition","level":2,"score":0.4415999948978424},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.42179998755455017},{"id":"https://openalex.org/C2985139394","wikidata":"https://www.wikidata.org/wiki/Q49908","display_name":"Sampling scheme","level":3,"score":0.4156999886035919},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.41280001401901245},{"id":"https://openalex.org/C198531522","wikidata":"https://www.wikidata.org/wiki/Q485146","display_name":"Sample (material)","level":2,"score":0.40400001406669617},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.40310001373291016},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.38019999861717224},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.3765000104904175},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.3671000003814697},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.35440000891685486},{"id":"https://openalex.org/C129848803","wikidata":"https://www.wikidata.org/wiki/Q2564360","display_name":"Sample size determination","level":2,"score":0.3199999928474426},{"id":"https://openalex.org/C122123141","wikidata":"https://www.wikidata.org/wiki/Q176623","display_name":"Random variable","level":2,"score":0.31290000677108765},{"id":"https://openalex.org/C2777317252","wikidata":"https://www.wikidata.org/wiki/Q18393516","display_name":"Rare events","level":2,"score":0.29670000076293945},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.27959999442100525},{"id":"https://openalex.org/C8272713","wikidata":"https://www.wikidata.org/wiki/Q176737","display_name":"Stochastic process","level":2,"score":0.27869999408721924},{"id":"https://openalex.org/C181243257","wikidata":"https://www.wikidata.org/wiki/Q1693522","display_name":"Sample mean and sample covariance","level":3,"score":0.2669000029563904},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.266400009393692},{"id":"https://openalex.org/C23130292","wikidata":"https://www.wikidata.org/wiki/Q5275358","display_name":"Differential privacy","level":2,"score":0.2639999985694885},{"id":"https://openalex.org/C72659945","wikidata":"https://www.wikidata.org/wiki/Q1503574","display_name":"Shot noise","level":3,"score":0.25870001316070557},{"id":"https://openalex.org/C166144826","wikidata":"https://www.wikidata.org/wiki/Q1145117","display_name":"Poisson process","level":3,"score":0.25130000710487366}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.07072","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.07072","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,"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":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.07072","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.07072","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,"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"},"sustainable_development_goals":[{"display_name":"Peace, Justice and strong institutions","score":0.41148659586906433,"id":"https://metadata.un.org/sdg/16"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Poisson":[0,92,177],"subsampling":[1,93,178],"is":[2,48,95,116,227],"the":[3,101,113,154,180,190,201],"default":[4],"sampling":[5,206],"scheme":[6,74],"in":[7,33,75,80,179,213],"differentially":[8],"private":[9,187],"machine":[10],"learning,":[11],"largely":[12],"because":[13],"its":[14],"unstructured":[15],"randomness":[16,25,207],"yields":[17,209],"tractable":[18],"privacy":[19,63,89,211],"amplification":[20,90],"analyses.":[21,163],"Yet":[22],"this":[23,41,46,136],"same":[24],"introduces":[26],"substantial":[27],"participation":[28,125,131,216],"variance:":[29],"each":[30,77],"sample":[31,78],"appears":[32],"very":[34],"different":[35],"numbers":[36],"of":[37,61,85,100,157],"training":[38],"iterations.":[39,133],"In":[40],"work,":[42],"we":[43,142],"show":[44,172],"that":[45,67,112,173,204],"variance":[47,126],"not":[49,118],"merely":[50],"a":[51,58,72,82,144],"practical":[52,145,169,221],"artifact":[53],"to":[54,196],"be":[55,218],"tolerated,":[56],"but":[57,122],"fundamental":[59],"source":[60],"suboptimal":[62],"amplification.":[64],"We":[65],"prove":[66],"Balanced":[68],"Iteration":[69],"Subsampling":[70],"(BIS),":[71],"structured":[73,215],"which":[76,152],"participates":[79],"exactly":[81],"fixed":[83],"number":[84],"iterations,":[86],"achieves":[87],"stronger":[88,210],"than":[91,167],"and":[94,106,160,222],"optimal":[96],"at":[97,229],"both":[98,219],"extremes":[99],"noise":[102,192],"spectrum":[103],"($\u03c3\\to":[104],"0$":[105],"$\u03c3\\to":[107],"\\infty$).":[108],"Our":[109,225],"analysis":[110],"reveals":[111],"privacy-noise":[114],"tradeoff":[115],"governed":[117],"by":[119,123,194],"maximizing":[120],"randomness,":[121],"eliminating":[124],"while":[127],"preserving":[128],"uniform":[129],"marginal":[130],"across":[132,165],"To":[134],"translate":[135],"asymptotic":[137],"theory":[138],"into":[139],"finite-noise":[140],"guarantees,":[141],"introduce":[143],"near-exact":[146],"Monte":[147],"Carlo":[148],"accountant":[149],"for":[150,185],"BIS,":[151],"removes":[153],"analytical":[155],"slack":[156],"existing":[158],"RDP":[159],"composition-based":[161],"PLD":[162],"Evaluations":[164],"more":[166,205,220,223],"60":[168],"DP-SGD":[170],"configurations":[171],"BIS":[174],"consistently":[175],"outperforms":[176],"low-noise":[181],"regimes":[182],"most":[183],"relevant":[184],"high-utility":[186],"training,":[188],"reducing":[189],"required":[191],"multiplier":[193],"up":[195],"$9.6\\%$.":[197],"These":[198],"results":[199],"overturn":[200],"common":[202],"intuition":[203],"necessarily":[208],"amplification:":[212],"DP-SGD,":[214],"can":[217],"private.":[224],"implementation":[226],"available":[228],"https://github.com/dong-xin-ao-andy/bis-mc-accountant.":[230]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-12T00:00:00"}