Model privacy: a unified framework for understanding model stealing attacks and defences
Journal of the Royal Statistical Society Series B: Statistical Methodology · 12.11.2026 · 88 (5) · 1515-1534
DOI: 10.1093/jrsssb/qkag059 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of the Royal Statistical Society Series B: Statistical Methodology, online erschienen am 02.04.2026. Verfasst von Ganghua Wang und 2 weiteren Autor:innen. Die Arbeit nennt 42 Literaturangaben. Gefördert von: National Science Foundation CAREER Programme.
Abstract (Originalsprache)
The use of machine learning (ML) has become increasingly prevalent in various domains, highlighting the importance of understanding and ensuring its safety. One pressing concern is the vulnerability of ML applications to model stealing attacks. These attacks involve adversaries attempting to recover a learned model through limited query-response interactions, such as those found in cloud-based services or on-chip artificial intelligence interfaces. While existing literature proposes various attack and defence strategies, these often lack a theoretical foundation and standardized evaluation criteria. In response, this work presents a framework called ‘Model Privacy’, providing a foundation for comprehensively analyzing model stealing attacks and defences. We establish a rigorous formulation for the threat model and objectives, propose methods to quantify the goodness of attack and defence strategies, and analyse the fundamental tradeoffs between utility and privacy in ML models. Our developed theory offers valuable insights into enhancing the security of ML models, especially highlighting the importance of the query-dependent structure of perturbations for effective defences. We demonstrate the application of model privacy from the defender’s perspective through various learning scenarios. Extensive experiments corroborate the insights and the effectiveness of defence mechanisms developed under the proposed framework.
Förderung
- National Science Foundation CAREER Programme
Zitieren
APA
Wang, G., Yang, Y., & Ding, J. (2026). Model privacy: a unified framework for understanding model stealing attacks and defences. Journal of the Royal Statistical Society Series B: Statistical Methodology, 88(5), 1515-1534. https://doi.org/10.1093/jrsssb/qkag059
Vancouver
Wang G, Yang Y, Ding J. Model privacy: a unified framework for understanding model stealing attacks and defences. Journal of the Royal Statistical Society Series B: Statistical Methodology. 2026;88(5):1515-1534. doi:10.1093/jrsssb/qkag059
BibTeX
@article{Wang2026,
author = {Wang, Ganghua and Yang, Yuhong and Ding, Jie},
title = {Model privacy: a unified framework for understanding model stealing attacks and defences},
journal = {Journal of the Royal Statistical Society Series B: Statistical Methodology},
year = {2026},
volume = {88},
number = {5},
pages = {1515--1534},
doi = {10.1093/jrsssb/qkag059}
}
EUVIMED