Sélection de la langue

Gouvernement du Canada / Government of Canada

Recherche

Protecting bilateral privacy in Machine Learning-as-a-Service: a differential privacy based defense

Par Conseil national de recherches du Canada

DOITrouver le DOI : https://doi.org/10.1007/978-981-99-9785-5_17
AuteurRechercher : Identifiant ORCID : https://orcid.org/0000-0002-4939-1642; Rechercher : Identifiant ORCID : https://orcid.org/0000-0002-1784-6091; Rechercher : Identifiant ORCID : https://orcid.org/0000-0001-8916-6645; Rechercher : 1Identifiant ORCID : https://orcid.org/0000-0002-3460-6946
Affiliation
  1. Conseil national de recherches du Canada. Technologies numériques
FormatTexte, Chapitre de livre
ConférenceAIS&P, December 3-5, 2023, Guangzhou, China
SujetMachine Learning as a Service; bilateral privacy; privacy leakage; model extraction; differential privacy
Résumé
With the continuous promotion and deepened application of Machine Learning-as-a-Service (MLaaS) across various societal domains, its privacy problems occur frequently and receive more and more attention from researchers. However, existing research focuses only on the client-side query privacy problem or only focuses on the server-side model privacy problem, and lacks a simultaneous focus on bilateral privacy defense schemes. In this paper, we design privacy-preserving mechanisms based on differential privacy for the client and server side respectively for the first time. By injecting noise into query requests and model responses, both the client and server sides in MLaaS are privacy-protected. Experimental results also demonstrate the effectiveness of the proposed solution in ensuring accuracy and providing privacy protection for both the clients and servers in MLaaS.
Date de publication
Maison d’éditionSpringer Nature
Dans
  • Artificial Intelligence Security and Privacy: First International Conference on Artificial Intelligence Security and Privacy, AIS&P 2023 (Proceedings, Part I) : 237–252.
Série
  • Lecture Notes in Computer Science 14509.
Langueanglais
Publications évaluées par des pairsOui
Exporter la noticeExporter en format RIS
Signaler une correctionSignaler une correction (s'ouvre dans un nouvel onglet)
Identificateur de l’enregistrement02579be0-382c-454c-9051-d2a4345919fb
Enregistrement créé2024-02-27
Enregistrement modifié2024-02-28
Date de modification :