{"id":"https://openalex.org/W7220908264","doi":"https://doi.org/10.48550/arxiv.2610.10483","title":"Two-Level Softmax Sampling Done Right: Correcting Bias from Size Imbalance and Dispersion","display_name":"Two-Level Softmax Sampling Done Right: Correcting Bias from Size Imbalance and Dispersion","publication_year":2026,"publication_date":"2026-10-07","ids":{"openalex":"https://openalex.org/W7220908264","doi":"https://doi.org/10.48550/arxiv.2610.10483"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.10483","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.10483","pdf_url":"https://arxiv.org/pdf/2610.10483","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/2610.10483","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5004252064","display_name":"Walid Bendada","orcid":"https://orcid.org/0009-0001-7510-650X","observed_orcids":["https://orcid.org/0009-0001-7510-650X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bendada, Walid","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5078969242","display_name":"Guillaume Salha-Galvan","orcid":"https://orcid.org/0000-0002-2452-1041","observed_orcids":["https://orcid.org/0000-0002-2452-1041"]},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Salha-Galvan, Guillaume","raw_affiliation_strings":["SJTU Paris Elite Institute of Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"SJTU Paris Elite Institute of Technology","institution_ids":["https://openalex.org/I183067930"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"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":0.6509000062942505,"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":0.6509000062942505,"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":0.12759999930858612,"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/T10028","display_name":"Topic Modeling","score":0.04580000042915344,"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/softmax-approximation","display_name":"softmax approximation","score":0.6420000195503235},{"id":"https://openalex.org/keywords/sampling-bias","display_name":"sampling bias","score":0.8679999709129333},{"id":"https://openalex.org/keywords/cluster-size-imbalance","display_name":"cluster size imbalance","score":0.8420000076293945},{"id":"https://openalex.org/keywords/cluster-sampling","display_name":"cluster sampling","score":0.4819999933242798}],"concepts":[{"id":"https://openalex.org/C188441871","wikidata":"https://www.wikidata.org/wiki/Q7554146","display_name":"Softmax function","level":3,"score":0.8723000288009644},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.7549999952316284},{"id":"https://openalex.org/C183380357","wikidata":"https://www.wikidata.org/wiki/Q1776598","display_name":"Cluster sampling","level":3,"score":0.5600000023841858},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5109999775886536},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.5055000185966492},{"id":"https://openalex.org/C103278499","wikidata":"https://www.wikidata.org/wiki/Q254465","display_name":"Similarity (geometry)","level":3,"score":0.4855000078678131},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.4848000109195709},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.46239998936653137},{"id":"https://openalex.org/C164866538","wikidata":"https://www.wikidata.org/wiki/Q367351","display_name":"Cluster (spacecraft)","level":2,"score":0.4327999949455261},{"id":"https://openalex.org/C177562468","wikidata":"https://www.wikidata.org/wiki/Q182893","display_name":"Dispersion (optics)","level":2,"score":0.4154999852180481},{"id":"https://openalex.org/C75917345","wikidata":"https://www.wikidata.org/wiki/Q2725298","display_name":"Sampling bias","level":3,"score":0.40070000290870667},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.3862000107765198},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.3700999915599823},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.365200012922287},{"id":"https://openalex.org/C129848803","wikidata":"https://www.wikidata.org/wiki/Q2564360","display_name":"Sample size determination","level":2,"score":0.36480000615119934},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.34380000829696655},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.3386000096797943},{"id":"https://openalex.org/C82152865","wikidata":"https://www.wikidata.org/wiki/Q7208505","display_name":"Poisson sampling","level":5,"score":0.32899999618530273},{"id":"https://openalex.org/C167723999","wikidata":"https://www.wikidata.org/wiki/Q3773214","display_name":"Sampling distribution","level":2,"score":0.29269999265670776},{"id":"https://openalex.org/C75373757","wikidata":"https://www.wikidata.org/wiki/Q7410160","display_name":"Sampling design","level":3,"score":0.2881999909877777},{"id":"https://openalex.org/C110121322","wikidata":"https://www.wikidata.org/wiki/Q865811","display_name":"Distribution (mathematics)","level":2,"score":0.2881999909877777},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.2770000100135803},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.2768000066280365},{"id":"https://openalex.org/C121864883","wikidata":"https://www.wikidata.org/wiki/Q677916","display_name":"Statistical physics","level":1,"score":0.2563000023365021},{"id":"https://openalex.org/C192489979","wikidata":"https://www.wikidata.org/wiki/Q1502997","display_name":"Systematic sampling","level":2,"score":0.2540000081062317}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.10483","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.10483","pdf_url":"https://arxiv.org/pdf/2610.10483","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.2610.10483","doi":"https://doi.org/10.48550/arxiv.2610.10483","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.10483","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:2610.10483","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.10483","pdf_url":"https://arxiv.org/pdf/2610.10483","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/W7220908264.pdf","grobid_xml":"https://content.openalex.org/works/W7220908264.grobid-xml"},"referenced_works_count":7,"referenced_works":["https://openalex.org/W2908692794","https://openalex.org/W4406482936","https://openalex.org/W4415199709","https://openalex.org/W4416940551","https://openalex.org/W7080126371","https://openalex.org/W7092286881","https://openalex.org/W7138089354"],"related_works":["https://openalex.org/W2774769176","https://openalex.org/W2734427577","https://openalex.org/W4406482936","https://openalex.org/W2962998615","https://openalex.org/W2787233435","https://openalex.org/W3126908443","https://openalex.org/W3167827288","https://openalex.org/W2554727591","https://openalex.org/W2946273996","https://openalex.org/W2786456373"],"abstract_inverted_index":{"Sampling":[0],"from":[1,73],"a":[2,6,32,47],"softmax":[3,28,108],"distribution":[4],"is":[5,31],"fundamental":[7],"operation":[8],"in":[9,16,135,140],"machine":[10],"learning,":[11],"but":[12],"its":[13,62],"linear":[14],"complexity":[15],"the":[17,123],"number":[18],"of":[19,127,137],"items":[20,39],"makes":[21],"exact":[22],"sampling":[23,30,69,89,125],"impractical":[24],"at":[25],"scale.":[26],"Two-level":[27],"(2LS)":[29],"popular":[33],"alternative":[34],"enabling":[35],"sublinear-time":[36],"sampling.":[37],"Assuming":[38],"are":[40],"partitioned":[41],"into":[42],"clusters,":[43],"2LS":[44,64,92,98,139],"first":[45],"samples":[46],"cluster":[48,79],"and":[49,67,82,94,96,104],"then":[50],"an":[51],"item":[52],"within":[53],"it.":[54],"In":[55],"this":[56],"paper,":[57],"we":[58],"show":[59],"that,":[60],"despite":[61],"advantages,":[63],"introduces":[65],"systematic":[66],"undesirable":[68],"biases,":[70],"which":[71,100],"arise":[72],"misweighting":[74],"clusters":[75],"by":[76],"ignoring":[77],"both":[78],"size":[80],"imbalance":[81],"intra-cluster":[83],"similarity":[84],"dispersion.":[85],"We":[86,130],"propose":[87],"two":[88],"methods,":[90],"Size-Corrected":[91],"(S-2LS)":[93],"Size-":[95],"Dispersion-Corrected":[97],"(SD-2LS),":[99],"correct":[101],"these":[102],"biases":[103],"provide":[105],"provably":[106],"better":[107],"approximations":[109],"with":[110],"negligible":[111],"to":[112],"non-existent":[113],"computational":[114],"overhead.":[115],"In-depth":[116],"experiments":[117],"on":[118],"five":[119],"large-scale":[120],"datasets":[121],"validate":[122],"improved":[124],"properties":[126],"our":[128],"methods.":[129],"recommend":[131],"their":[132],"consistent":[133],"use":[134],"place":[136],"standard":[138],"future":[141],"work.":[142]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-09T00:00:00"}