Theoretically Unmasking Inference Attacks Against LDP-Protected Clients in Federated Vision Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Nguyen, Quan, Vu, Minh N., Nguyen, Truc, Thai, My T.
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866918109129998336
author Nguyen, Quan
Vu, Minh N.
Nguyen, Truc
Thai, My T.
author_facet Nguyen, Quan
Vu, Minh N.
Nguyen, Truc
Thai, My T.
contents Federated Learning enables collaborative learning among clients via a coordinating server while avoiding direct data sharing, offering a perceived solution to preserve privacy. However, recent studies on Membership Inference Attacks (MIAs) have challenged this notion, showing high success rates against unprotected training data. While local differential privacy (LDP) is widely regarded as a gold standard for privacy protection in data analysis, most studies on MIAs either neglect LDP or fail to provide theoretical guarantees for attack success rates against LDP-protected data. To address this gap, we derive theoretical lower bounds for the success rates of low-polynomial time MIAs that exploit vulnerabilities in fully connected or self-attention layers. We establish that even when data are protected by LDP, privacy risks persist, depending on the privacy budget. Practical evaluations on federated vision models confirm considerable privacy risks, revealing that the noise required to mitigate these attacks significantly degrades models' utility.
format Preprint
id arxiv_https___arxiv_org_abs_2506_17292
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Theoretically Unmasking Inference Attacks Against LDP-Protected Clients in Federated Vision Models
Nguyen, Quan
Vu, Minh N.
Nguyen, Truc
Thai, My T.
Cryptography and Security
Artificial Intelligence
Federated Learning enables collaborative learning among clients via a coordinating server while avoiding direct data sharing, offering a perceived solution to preserve privacy. However, recent studies on Membership Inference Attacks (MIAs) have challenged this notion, showing high success rates against unprotected training data. While local differential privacy (LDP) is widely regarded as a gold standard for privacy protection in data analysis, most studies on MIAs either neglect LDP or fail to provide theoretical guarantees for attack success rates against LDP-protected data. To address this gap, we derive theoretical lower bounds for the success rates of low-polynomial time MIAs that exploit vulnerabilities in fully connected or self-attention layers. We establish that even when data are protected by LDP, privacy risks persist, depending on the privacy budget. Practical evaluations on federated vision models confirm considerable privacy risks, revealing that the noise required to mitigate these attacks significantly degrades models' utility.
title Theoretically Unmasking Inference Attacks Against LDP-Protected Clients in Federated Vision Models
topic Cryptography and Security
Artificial Intelligence
url https://arxiv.org/abs/2506.17292