LDPKiT: Superimposing Remote Queries for Privacy-Preserving Local Model Training Article Swipe

PDF

Kexin Li , Xi Yang , Aastha Mehta , David Lie ·

YOU? · · 2024 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2405.16361

Users of modern Machine Learning (ML) cloud services face a privacy conundrum -- on one hand, they may have concerns about sending private data to the service for inference, but on the other hand, for specialized models, there may be no alternative but to use the proprietary model of the ML service. In this work, we present LDPKiT, a framework for non-adversarial, privacy-preserving model extraction that leverages a user's private in-distribution data while bounding privacy leakage. LDPKiT introduces a novel superimposition technique that generates approximately in-distribution samples, enabling effective knowledge transfer under local differential privacy (LDP). Experiments on Fashion-MNIST, SVHN, and PathMNIST demonstrate that LDPKiT consistently improves utility while maintaining privacy, with benefits that become more pronounced at stronger noise levels. For example, on SVHN, LDPKiT achieves nearly the same inference accuracy at $ε=1.25$ as at $ε=2.0$, yielding stronger privacy guarantees with less than a 2% accuracy reduction. We further conduct sensitivity analyses to examine the effect of dataset size on performance and provide a systematic analysis of latent space representations, offering theoretical insights into the accuracy gains of LDPKiT.

Related Topics

Computer Science

Artificial Intelligence

Geography

Meteorology

Concepts

Training (meteorology) Computer science Noise (video) Artificial intelligence Image (mathematics) Geography Meteorology

Metadata

Type: preprint
Language: en
Landing Page: http://arxiv.org/abs/2405.16361
PDF: https://arxiv.org/pdf/2405.16361
OA Status: green
Related Works: 10
OpenAlex ID: https://openalex.org/W4399115546

All OpenAlex metadata

Raw OpenAlex JSON

OpenAlex ID: https://openalex.org/W4399115546

Canonical identifier for this work in OpenAlex
DOI: https://doi.org/10.48550/arxiv.2405.16361

Digital Object Identifier
Title: LDPKiT: Superimposing Remote Queries for Privacy-Preserving Local Model Training

Work title
Type: preprint

OpenAlex work type
Language: en

Primary language
Publication year: 2024

Year of publication
Publication date: 2024-05-25

Full publication date if available
Authors: Kexin Li, Xi Yang, Aastha Mehta, David Lie

List of authors in order
Landing page: https://arxiv.org/abs/2405.16361

Publisher landing page
PDF URL: https://arxiv.org/pdf/2405.16361

Direct link to full text PDF
Open access: Yes

Whether a free full text is available
OA status: green

Open access status per OpenAlex
OA URL: https://arxiv.org/pdf/2405.16361

Direct OA link when available
Concepts: Training (meteorology), Computer science, Noise (video), Artificial intelligence, Image (mathematics), Geography, Meteorology

Top concepts (fields/topics) attached by OpenAlex
Cited by: 0

Total citation count in OpenAlex
Related works (count): 10

Other works algorithmically related by OpenAlex

Full payload

id	https://openalex.org/W4399115546
doi	https://doi.org/10.48550/arxiv.2405.16361
ids.doi	https://doi.org/10.48550/arxiv.2405.16361
ids.openalex	https://openalex.org/W4399115546
fwci
type	preprint
title	LDPKiT: Superimposing Remote Queries for Privacy-Preserving Local Model Training
biblio.issue
biblio.volume
biblio.last_page
biblio.first_page
topics[0].id	https://openalex.org/T10232
topics[0].field.id	https://openalex.org/fields/22
topics[0].field.display_name	Engineering
topics[0].score	0.9815000295639038
topics[0].domain.id	https://openalex.org/domains/3
topics[0].domain.display_name	Physical Sciences
topics[0].subfield.id	https://openalex.org/subfields/2208
topics[0].subfield.display_name	Electrical and Electronic Engineering
topics[0].display_name	Optical Network Technologies
topics[1].id	https://openalex.org/T10847
topics[1].field.id	https://openalex.org/fields/22
topics[1].field.display_name	Engineering
topics[1].score	0.975600004196167
topics[1].domain.id	https://openalex.org/domains/3
topics[1].domain.display_name	Physical Sciences
topics[1].subfield.id	https://openalex.org/subfields/2208
topics[1].subfield.display_name	Electrical and Electronic Engineering
topics[1].display_name	Advanced Optical Network Technologies
topics[2].id	https://openalex.org/T10767
topics[2].field.id	https://openalex.org/fields/22
topics[2].field.display_name	Engineering
topics[2].score	0.9297000169754028
topics[2].domain.id	https://openalex.org/domains/3
topics[2].domain.display_name	Physical Sciences
topics[2].subfield.id	https://openalex.org/subfields/2208
topics[2].subfield.display_name	Electrical and Electronic Engineering
topics[2].display_name	Advanced Photonic Communication Systems
is_xpac	False
apc_list
apc_paid
concepts[0].id	https://openalex.org/C2777211547
concepts[0].level	2
concepts[0].score	0.7120229601860046
concepts[0].wikidata	https://www.wikidata.org/wiki/Q17141490
concepts[0].display_name	Training (meteorology)
concepts[1].id	https://openalex.org/C41008148
concepts[1].level	0
concepts[1].score	0.6542249321937561
concepts[1].wikidata	https://www.wikidata.org/wiki/Q21198
concepts[1].display_name	Computer science
concepts[2].id	https://openalex.org/C99498987
concepts[2].level	3
concepts[2].score	0.5654084086418152
concepts[2].wikidata	https://www.wikidata.org/wiki/Q2210247
concepts[2].display_name	Noise (video)
concepts[3].id	https://openalex.org/C154945302
concepts[3].level	1
concepts[3].score	0.30991941690444946
concepts[3].wikidata	https://www.wikidata.org/wiki/Q11660
concepts[3].display_name	Artificial intelligence
concepts[4].id	https://openalex.org/C115961682
concepts[4].level	2
concepts[4].score	0.07172545790672302
concepts[4].wikidata	https://www.wikidata.org/wiki/Q860623
concepts[4].display_name	Image (mathematics)
concepts[5].id	https://openalex.org/C205649164
concepts[5].level	0
concepts[5].score	0.0647808313369751
concepts[5].wikidata	https://www.wikidata.org/wiki/Q1071
concepts[5].display_name	Geography
concepts[6].id	https://openalex.org/C153294291
concepts[6].level	1
concepts[6].score	0.0
concepts[6].wikidata	https://www.wikidata.org/wiki/Q25261
concepts[6].display_name	Meteorology
keywords[0].id	https://openalex.org/keywords/training
keywords[0].score	0.7120229601860046
keywords[0].display_name	Training (meteorology)
keywords[1].id	https://openalex.org/keywords/computer-science
keywords[1].score	0.6542249321937561
keywords[1].display_name	Computer science
keywords[2].id	https://openalex.org/keywords/noise
keywords[2].score	0.5654084086418152
keywords[2].display_name	Noise (video)
keywords[3].id	https://openalex.org/keywords/artificial-intelligence
keywords[3].score	0.30991941690444946
keywords[3].display_name	Artificial intelligence
keywords[4].id	https://openalex.org/keywords/image
keywords[4].score	0.07172545790672302
keywords[4].display_name	Image (mathematics)
keywords[5].id	https://openalex.org/keywords/geography
keywords[5].score	0.0647808313369751
keywords[5].display_name	Geography
language	en
locations[0].id	pmh:oai:arXiv.org:2405.16361
locations[0].is_oa	True
locations[0].source.id	https://openalex.org/S4306400194
locations[0].source.issn
locations[0].source.type	repository
locations[0].source.is_oa	True
locations[0].source.issn_l
locations[0].source.is_core	False
locations[0].source.is_in_doaj	False
locations[0].source.display_name	arXiv (Cornell University)
locations[0].source.host_organization	https://openalex.org/I205783295
locations[0].source.host_organization_name	Cornell University
locations[0].source.host_organization_lineage	https://openalex.org/I205783295
locations[0].license
locations[0].pdf_url	https://arxiv.org/pdf/2405.16361
locations[0].version	submittedVersion
locations[0].raw_type
locations[0].license_id
locations[0].is_accepted	False
locations[0].is_published	False
locations[0].raw_source_name
locations[0].landing_page_url	http://arxiv.org/abs/2405.16361
locations[1].id	doi:10.48550/arxiv.2405.16361
locations[1].is_oa	True
locations[1].source.id	https://openalex.org/S4306400194
locations[1].source.issn
locations[1].source.type	repository
locations[1].source.is_oa	True
locations[1].source.issn_l
locations[1].source.is_core	False
locations[1].source.is_in_doaj	False
locations[1].source.display_name	arXiv (Cornell University)
locations[1].source.host_organization	https://openalex.org/I205783295
locations[1].source.host_organization_name	Cornell University
locations[1].source.host_organization_lineage	https://openalex.org/I205783295
locations[1].license	cc-by
locations[1].pdf_url
locations[1].version
locations[1].raw_type	article
locations[1].license_id	https://openalex.org/licenses/cc-by
locations[1].is_accepted	False
locations[1].is_published
locations[1].raw_source_name
locations[1].landing_page_url	https://doi.org/10.48550/arxiv.2405.16361
indexed_in	arxiv, datacite
authorships[0].author.id	https://openalex.org/A5111114874
authorships[0].author.orcid	https://orcid.org/0009-0008-7417-6398
authorships[0].author.display_name	Kexin Li
authorships[0].author_position	first
authorships[0].raw_author_name	Li, Kexin
authorships[0].is_corresponding	False
authorships[1].author.id	https://openalex.org/A5054583558
authorships[1].author.orcid	https://orcid.org/0000-0002-3359-4285
authorships[1].author.display_name	Xi Yang
authorships[1].author_position	middle
authorships[1].raw_author_name	Xi, Yang
authorships[1].is_corresponding	False
authorships[2].author.id	https://openalex.org/A5086794440
authorships[2].author.orcid	https://orcid.org/0009-0005-3416-5254
authorships[2].author.display_name	Aastha Mehta
authorships[2].author_position	middle
authorships[2].raw_author_name	Mehta, Aastha
authorships[2].is_corresponding	False
authorships[3].author.id	https://openalex.org/A5049933072
authorships[3].author.orcid	https://orcid.org/0000-0002-2000-6827
authorships[3].author.display_name	David Lie
authorships[3].author_position	last
authorships[3].raw_author_name	Lie, David
authorships[3].is_corresponding	False
has_content.pdf	False
has_content.grobid_xml	False
is_paratext	False
open_access.is_oa	True
open_access.oa_url	https://arxiv.org/pdf/2405.16361
open_access.oa_status	green
open_access.any_repository_has_fulltext	False
created_date	2024-05-29T00:00:00
display_name	LDPKiT: Superimposing Remote Queries for Privacy-Preserving Local Model Training
has_fulltext	False
is_retracted	False
updated_date	2025-11-06T06:51:31.235846
primary_topic.id	https://openalex.org/T10232
primary_topic.field.id	https://openalex.org/fields/22
primary_topic.field.display_name	Engineering
primary_topic.score	0.9815000295639038
primary_topic.domain.id	https://openalex.org/domains/3
primary_topic.domain.display_name	Physical Sciences
primary_topic.subfield.id	https://openalex.org/subfields/2208
primary_topic.subfield.display_name	Electrical and Electronic Engineering
primary_topic.display_name	Optical Network Technologies
related_works	https://openalex.org/W4391375266, https://openalex.org/W2748952813, https://openalex.org/W230091440, https://openalex.org/W2390279801, https://openalex.org/W2233261550, https://openalex.org/W2358668433, https://openalex.org/W4396701345, https://openalex.org/W2810751659, https://openalex.org/W258997015, https://openalex.org/W2376932109
cited_by_count	0
locations_count	2
best_oa_location.id	pmh:oai:arXiv.org:2405.16361
best_oa_location.is_oa	True
best_oa_location.source.id	https://openalex.org/S4306400194
best_oa_location.source.issn
best_oa_location.source.type	repository
best_oa_location.source.is_oa	True
best_oa_location.source.issn_l
best_oa_location.source.is_core	False
best_oa_location.source.is_in_doaj	False
best_oa_location.source.display_name	arXiv (Cornell University)
best_oa_location.source.host_organization	https://openalex.org/I205783295
best_oa_location.source.host_organization_name	Cornell University
best_oa_location.source.host_organization_lineage	https://openalex.org/I205783295
best_oa_location.license
best_oa_location.pdf_url	https://arxiv.org/pdf/2405.16361
best_oa_location.version	submittedVersion
best_oa_location.raw_type
best_oa_location.license_id
best_oa_location.is_accepted	False
best_oa_location.is_published	False
best_oa_location.raw_source_name
best_oa_location.landing_page_url	http://arxiv.org/abs/2405.16361
primary_location.id	pmh:oai:arXiv.org:2405.16361
primary_location.is_oa	True
primary_location.source.id	https://openalex.org/S4306400194
primary_location.source.issn
primary_location.source.type	repository
primary_location.source.is_oa	True
primary_location.source.issn_l
primary_location.source.is_core	False
primary_location.source.is_in_doaj	False
primary_location.source.display_name	arXiv (Cornell University)
primary_location.source.host_organization	https://openalex.org/I205783295
primary_location.source.host_organization_name	Cornell University
primary_location.source.host_organization_lineage	https://openalex.org/I205783295
primary_location.license
primary_location.pdf_url	https://arxiv.org/pdf/2405.16361
primary_location.version	submittedVersion
primary_location.raw_type
primary_location.license_id
primary_location.is_accepted	False
primary_location.is_published	False
primary_location.raw_source_name
primary_location.landing_page_url	http://arxiv.org/abs/2405.16361
publication_date	2024-05-25
publication_year	2024
referenced_works_count	0
abstract_inverted_index.a	9, 58, 67, 78, 144, 164
abstract_inverted_index.--	12
abstract_inverted_index.2%	145
abstract_inverted_index.In	52
abstract_inverted_index.ML	50
abstract_inverted_index.We	148
abstract_inverted_index.as	134
abstract_inverted_index.at	117, 132, 135
abstract_inverted_index.be	39
abstract_inverted_index.no	40
abstract_inverted_index.of	1, 48, 157, 167, 178
abstract_inverted_index.on	13, 30, 97, 123, 160
abstract_inverted_index.to	24, 43, 153
abstract_inverted_index.we	55
abstract_inverted_index.For	121
abstract_inverted_index.and	100, 162
abstract_inverted_index.but	29, 42
abstract_inverted_index.for	27, 34, 60
abstract_inverted_index.may	17, 38
abstract_inverted_index.one	14
abstract_inverted_index.the	25, 31, 45, 49, 128, 155, 175
abstract_inverted_index.use	44
abstract_inverted_index.(ML)	5
abstract_inverted_index.data	23, 71
abstract_inverted_index.face	8
abstract_inverted_index.have	18
abstract_inverted_index.into	174
abstract_inverted_index.less	142
abstract_inverted_index.more	115
abstract_inverted_index.same	129
abstract_inverted_index.size	159
abstract_inverted_index.than	143
abstract_inverted_index.that	65, 82, 103, 113
abstract_inverted_index.they	16
abstract_inverted_index.this	53
abstract_inverted_index.with	111, 141
abstract_inverted_index.SVHN,	99, 124
abstract_inverted_index.Users	0
abstract_inverted_index.about	20
abstract_inverted_index.cloud	6
abstract_inverted_index.gains	177
abstract_inverted_index.hand,	15, 33
abstract_inverted_index.local	92
abstract_inverted_index.model	47, 63
abstract_inverted_index.noise	119
abstract_inverted_index.novel	79
abstract_inverted_index.other	32
abstract_inverted_index.space	169
abstract_inverted_index.there	37
abstract_inverted_index.under	91
abstract_inverted_index.while	72, 108
abstract_inverted_index.work,	54
abstract_inverted_index.(LDP).	95
abstract_inverted_index.LDPKiT	76, 104, 125
abstract_inverted_index.become	114
abstract_inverted_index.effect	156
abstract_inverted_index.latent	168
abstract_inverted_index.modern	2
abstract_inverted_index.nearly	127
abstract_inverted_index.user's	68
abstract_inverted_index.LDPKiT,	57
abstract_inverted_index.LDPKiT.	179
abstract_inverted_index.Machine	3
abstract_inverted_index.conduct	150
abstract_inverted_index.dataset	158
abstract_inverted_index.examine	154
abstract_inverted_index.further	149
abstract_inverted_index.levels.	120
abstract_inverted_index.models,	36
abstract_inverted_index.present	56
abstract_inverted_index.privacy	10, 74, 94, 139
abstract_inverted_index.private	22, 69
abstract_inverted_index.provide	163
abstract_inverted_index.sending	21
abstract_inverted_index.service	26
abstract_inverted_index.utility	107
abstract_inverted_index.Learning	4
abstract_inverted_index.accuracy	131, 146, 176
abstract_inverted_index.achieves	126
abstract_inverted_index.analyses	152
abstract_inverted_index.analysis	166
abstract_inverted_index.benefits	112
abstract_inverted_index.bounding	73
abstract_inverted_index.concerns	19
abstract_inverted_index.enabling	87
abstract_inverted_index.example,	122
abstract_inverted_index.improves	106
abstract_inverted_index.insights	173
abstract_inverted_index.leakage.	75
abstract_inverted_index.offering	171
abstract_inverted_index.privacy,	110
abstract_inverted_index.samples,	86
abstract_inverted_index.service.	51
abstract_inverted_index.services	7
abstract_inverted_index.stronger	118, 138
abstract_inverted_index.transfer	90
abstract_inverted_index.yielding	137
abstract_inverted_index.$ε=1.25$	133
abstract_inverted_index.$ε=2.0$,	136
abstract_inverted_index.PathMNIST	101
abstract_inverted_index.conundrum	11
abstract_inverted_index.effective	88
abstract_inverted_index.framework	59
abstract_inverted_index.generates	83
abstract_inverted_index.inference	130
abstract_inverted_index.knowledge	89
abstract_inverted_index.leverages	66
abstract_inverted_index.technique	81
abstract_inverted_index.extraction	64
abstract_inverted_index.guarantees	140
abstract_inverted_index.inference,	28
abstract_inverted_index.introduces	77
abstract_inverted_index.pronounced	116
abstract_inverted_index.reduction.	147
abstract_inverted_index.systematic	165
abstract_inverted_index.Experiments	96
abstract_inverted_index.alternative	41
abstract_inverted_index.demonstrate	102
abstract_inverted_index.maintaining	109
abstract_inverted_index.performance	161
abstract_inverted_index.proprietary	46
abstract_inverted_index.sensitivity	151
abstract_inverted_index.specialized	35
abstract_inverted_index.theoretical	172
abstract_inverted_index.consistently	105
abstract_inverted_index.differential	93
abstract_inverted_index.approximately	84
abstract_inverted_index.Fashion-MNIST,	98
abstract_inverted_index.in-distribution	70, 85
abstract_inverted_index.superimposition	80
abstract_inverted_index.non-adversarial,	61
abstract_inverted_index.representations,	170
abstract_inverted_index.privacy-preserving	62
cited_by_percentile_year
countries_distinct_count	0
institutions_distinct_count	4
citation_normalized_percentile