{"id":"https://openalex.org/W7214746261","doi":"https://doi.org/10.48550/arxiv.2609.32962","title":"Clipped or Unclipped? Finite-Sample Trade-offs for Averaged SGD under Heavy-Tailed Noise","display_name":"Clipped or Unclipped? Finite-Sample Trade-offs for Averaged SGD under Heavy-Tailed Noise","publication_year":2026,"publication_date":"2026-09-26","ids":{"openalex":"https://openalex.org/W7214746261","doi":"https://doi.org/10.48550/arxiv.2609.32962"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.32962","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.32962","pdf_url":"https://arxiv.org/pdf/2609.32962","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/2609.32962","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5013083165","display_name":"Alexandra Suvorikova","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210116741","display_name":"Innopolis University","ror":"https://ror.org/02b7jh107","country_code":"RU","type":"education","lineage":["https://openalex.org/I4210116741"]}],"countries":["RU"],"is_corresponding":false,"raw_author_name":"Suvorikova, Alexandra","raw_affiliation_strings":["Innopolis University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Innopolis University","institution_ids":["https://openalex.org/I4210116741"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5026215052","display_name":"Egor Leonidovich Gladin","orcid":"https://orcid.org/0000-0002-9086-996X","observed_orcids":["https://orcid.org/0000-0002-9086-996X"]},"institutions":[{"id":"https://openalex.org/I118501908","display_name":"National Research University Higher School of Economics","ror":"https://ror.org/055f7t516","country_code":"RU","type":"education","lineage":["https://openalex.org/I118501908"]}],"countries":["RU"],"is_corresponding":false,"raw_author_name":"Gladin, Egor","raw_affiliation_strings":["HSE University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"HSE University","institution_ids":["https://openalex.org/I118501908"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5000974070","display_name":"Darina Dvinskikh","orcid":"https://orcid.org/0000-0003-1757-1021","observed_orcids":["https://orcid.org/0000-0003-1757-1021"]},"institutions":[{"id":"https://openalex.org/I118501908","display_name":"National Research University Higher School of Economics","ror":"https://ror.org/055f7t516","country_code":"RU","type":"education","lineage":["https://openalex.org/I118501908"]}],"countries":["RU"],"is_corresponding":false,"raw_author_name":"Dvinskikh, Darina","raw_affiliation_strings":["HSE University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"HSE University","institution_ids":["https://openalex.org/I118501908"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5033094238","display_name":"Artem Agafonov","orcid":"https://orcid.org/0000-0001-9314-8634","observed_orcids":["https://orcid.org/0000-0001-9314-8634"]},"institutions":[{"id":"https://openalex.org/I4210116741","display_name":"Innopolis University","ror":"https://ror.org/02b7jh107","country_code":"RU","type":"education","lineage":["https://openalex.org/I4210116741"]}],"countries":["RU"],"is_corresponding":false,"raw_author_name":"Agafonov, Artem","raw_affiliation_strings":["Innopolis University MIRAI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Innopolis University MIRAI","institution_ids":["https://openalex.org/I4210116741"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5030490030","display_name":"Mohammad Soud Alkousa","orcid":"https://orcid.org/0000-0001-5470-0182","observed_orcids":["https://orcid.org/0000-0001-5470-0182"]},"institutions":[{"id":"https://openalex.org/I4210116741","display_name":"Innopolis University","ror":"https://ror.org/02b7jh107","country_code":"RU","type":"education","lineage":["https://openalex.org/I4210116741"]}],"countries":["RU"],"is_corresponding":false,"raw_author_name":"Alkousa, Mohammad","raw_affiliation_strings":["Innopolis University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Innopolis University","institution_ids":["https://openalex.org/I4210116741"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5069069110","display_name":"Yuriy  V. Dorn","orcid":"https://orcid.org/0000-0003-0533-3018","observed_orcids":["https://orcid.org/0000-0003-0533-3018"]},"institutions":[{"id":"https://openalex.org/I19880235","display_name":"Lomonosov Moscow State University","ror":"https://ror.org/010pmpe69","country_code":"RU","type":"education","lineage":["https://openalex.org/I19880235"]}],"countries":["RU"],"is_corresponding":false,"raw_author_name":"Dorn, Yuriy","raw_affiliation_strings":["MSU AI Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"MSU AI Institute","institution_ids":["https://openalex.org/I19880235"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5044833995","display_name":"Vladislav Vyacheslavovich Matyukhin","orcid":"https://orcid.org/0000-0003-4961-1577","observed_orcids":["https://orcid.org/0000-0003-4961-1577"]},"institutions":[{"id":"https://openalex.org/I4210116741","display_name":"Innopolis University","ror":"https://ror.org/02b7jh107","country_code":"RU","type":"education","lineage":["https://openalex.org/I4210116741"]}],"countries":["RU"],"is_corresponding":false,"raw_author_name":"Matyukhin, Vladislav","raw_affiliation_strings":["Innopolis University MIRAI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Innopolis University MIRAI","institution_ids":["https://openalex.org/I4210116741"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5028866398","display_name":"Alexander  Vladimirovich Gasnikov","orcid":"https://orcid.org/0000-0002-7386-039X","observed_orcids":["https://orcid.org/0000-0002-7386-039X"]},"institutions":[{"id":"https://openalex.org/I4210116741","display_name":"Innopolis University","ror":"https://ror.org/02b7jh107","country_code":"RU","type":"education","lineage":["https://openalex.org/I4210116741"]}],"countries":["RU"],"is_corresponding":false,"raw_author_name":"Gasnikov, Alexander","raw_affiliation_strings":["Innopolis University MIRAI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Innopolis University MIRAI","institution_ids":["https://openalex.org/I4210116741"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"corresponding_author_ids":[],"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.9994999766349792,"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":0.9994999766349792,"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":0.0,"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/T10963","display_name":"Advanced Optimization Algorithms Research","score":0.0,"subfield":{"id":"https://openalex.org/subfields/2612","display_name":"Numerical Analysis"},"field":{"id":"https://openalex.org/fields/26","display_name":"Mathematics"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/gradient-clipping","display_name":"gradient clipping","score":0.9959999918937683},{"id":"https://openalex.org/keywords/polyak-ruppert-averaging","display_name":"Polyak-Ruppert averaging","score":0.9459999799728394},{"id":"https://openalex.org/keywords/heavy-tailed-noise","display_name":"heavy-tailed noise","score":0.9810000061988831},{"id":"https://openalex.org/keywords/finite-sample-analysis","display_name":"finite-sample analysis","score":0.7940000295639038},{"id":"https://openalex.org/keywords/confidence-limits","display_name":"confidence limits","score":0.6000000238418579},{"id":"https://openalex.org/keywords/asymptotic-variance","display_name":"asymptotic variance","score":0.7400000095367432},{"id":"https://openalex.org/keywords/asymmetric-noise","display_name":"asymmetric noise","score":0.6349999904632568}],"concepts":[{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.7914999723434448},{"id":"https://openalex.org/C2776848632","wikidata":"https://www.wikidata.org/wiki/Q853463","display_name":"Clipping (morphology)","level":2,"score":0.6053000092506409},{"id":"https://openalex.org/C163716315","wikidata":"https://www.wikidata.org/wiki/Q901177","display_name":"Gaussian","level":2,"score":0.5246999859809875},{"id":"https://openalex.org/C129844170","wikidata":"https://www.wikidata.org/wiki/Q41299","display_name":"Quadratic equation","level":2,"score":0.5080000162124634},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.4814000129699707},{"id":"https://openalex.org/C57691317","wikidata":"https://www.wikidata.org/wiki/Q1289248","display_name":"Scalar (mathematics)","level":2,"score":0.4253000020980835},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.4212999939918518},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.4009000062942505},{"id":"https://openalex.org/C168773036","wikidata":"https://www.wikidata.org/wiki/Q264164","display_name":"Recursion (computer science)","level":2,"score":0.3864000141620636},{"id":"https://openalex.org/C207467116","wikidata":"https://www.wikidata.org/wiki/Q4385666","display_name":"Inverse","level":2,"score":0.37229999899864197},{"id":"https://openalex.org/C77553402","wikidata":"https://www.wikidata.org/wiki/Q13222579","display_name":"Upper and lower bounds","level":2,"score":0.3596999943256378},{"id":"https://openalex.org/C4199805","wikidata":"https://www.wikidata.org/wiki/Q2725903","display_name":"Gaussian noise","level":2,"score":0.3508000075817108},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.3402999937534332},{"id":"https://openalex.org/C58315980","wikidata":"https://www.wikidata.org/wiki/Q179436","display_name":"Asymptote","level":2,"score":0.32670000195503235},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.303600013256073},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.30239999294281006},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.28839999437332153},{"id":"https://openalex.org/C151201525","wikidata":"https://www.wikidata.org/wiki/Q177239","display_name":"Limit (mathematics)","level":2,"score":0.28200000524520874},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2793999910354614},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.27799999713897705},{"id":"https://openalex.org/C51267290","wikidata":"https://www.wikidata.org/wiki/Q5527848","display_name":"Gaussian random field","level":4,"score":0.27379998564720154},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.26829999685287476},{"id":"https://openalex.org/C97256817","wikidata":"https://www.wikidata.org/wiki/Q1462316","display_name":"Spurious relationship","level":2,"score":0.2671000063419342},{"id":"https://openalex.org/C2986577269","wikidata":"https://www.wikidata.org/wiki/Q11306265","display_name":"Random noise","level":2,"score":0.25929999351501465},{"id":"https://openalex.org/C112680207","wikidata":"https://www.wikidata.org/wiki/Q714886","display_name":"Regular polygon","level":2,"score":0.25679999589920044},{"id":"https://openalex.org/C47446073","wikidata":"https://www.wikidata.org/wiki/Q5165890","display_name":"Control theory (sociology)","level":3,"score":0.25189998745918274},{"id":"https://openalex.org/C122048520","wikidata":"https://www.wikidata.org/wiki/Q2913954","display_name":"Percentile","level":2,"score":0.25110000371932983}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.32962","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.32962","pdf_url":"https://arxiv.org/pdf/2609.32962","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.2609.32962","doi":"https://doi.org/10.48550/arxiv.2609.32962","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.32962","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:2609.32962","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.32962","pdf_url":"https://arxiv.org/pdf/2609.32962","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/W7214746261.pdf","grobid_xml":"https://content.openalex.org/works/W7214746261.grobid-xml"},"referenced_works_count":8,"referenced_works":["https://openalex.org/W2905800572","https://openalex.org/W3126750930","https://openalex.org/W4320721098","https://openalex.org/W4384918448","https://openalex.org/W4417293761","https://openalex.org/W6948236204","https://openalex.org/W7090864657","https://openalex.org/W7163157403"],"related_works":["https://openalex.org/W4415036353","https://openalex.org/W7223094074","https://openalex.org/W4320721098","https://openalex.org/W7160275576","https://openalex.org/W4378588713","https://openalex.org/W4362655248","https://openalex.org/W4415798624","https://openalex.org/W3026800838","https://openalex.org/W7221794827","https://openalex.org/W4403580163"],"abstract_inverted_index":{"Gradient":[0],"clipping":[1,63,72,149],"is":[2],"widely":[3],"used":[4],"to":[5,124],"stabilize":[6],"training,":[7],"but":[8],"it":[9],"need":[10,64],"not":[11,65],"improve":[12,66],"the":[13,54,58,82,91,99,103,110,122,152],"statistical":[14],"accuracy":[15,46],"of":[16,28,102,109,133],"averaged":[17,33],"SGD,":[18],"even":[19],"under":[20,35,50,155],"heavy-tailed":[21],"noise.":[22,157],"We":[23,106,128],"derive":[24],"a":[25,78,143],"finite-sample":[26],"comparison":[27,123],"clipped":[29],"and":[30,47,74,120],"unclipped":[31,59,111],"Polyak-Ruppert":[32],"SGD":[34],"finite":[36,137],"conditional":[37],"$p$-th":[38],"moments,":[39],"$p\\ge2$.":[40],"Our":[41],"main":[42],"result":[43],"gives":[44],"explicit":[45],"confidence":[48,100],"conditions":[49],"which,":[51],"for":[52],"$p>2$,":[53],"Gaussian":[55,145],"term":[56,113],"dominates":[57],"deviation":[60],"bound,":[61],"so":[62],"its":[67],"leading":[68,104],"order.":[69],"By":[70],"balancing":[71],"bias":[73],"concentration,":[75],"we":[76],"obtain":[77],"bound":[79],"in":[80],"which":[81],"heavy-tail":[83,112],"correction":[84],"depends":[85],"logarithmically":[86],"rather":[87],"than":[88],"polynomially":[89],"on":[90,142],"inverse":[92],"failure":[93],"probability.":[94],"At":[95],"$p=2$,":[96],"this":[97],"improves":[98],"dependence":[101],"bound.":[105],"establish":[107],"sharpness":[108],"through":[114],"an":[115],"exact":[116],"one-dimensional":[117],"quadratic":[118],"recursion":[119],"extend":[121],"projected":[125],"convex":[126],"SGD.":[127],"also":[129],"prove":[130],"concrete":[131],"costs":[132],"clipping:":[134],"every":[135],"fixed":[136],"threshold":[138],"increases":[139],"asymptotic":[140],"variance":[141],"scalar":[144],"quadratic,":[146],"while":[147],"whole-gradient":[148],"can":[150],"shift":[151],"limiting":[153],"point":[154],"asymmetric":[156]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-30T00:00:00"}