发表机构
Dauphine PSL; CNRS（巴黎第九大学PSL校区; 法国国家科学研究中心）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究结合表演式学习与隐私保护，提出表演式隐私概念，通过理论与实验证明，当数据泄露与用户参与的反馈回路足够强时，有限隐私预算的差分隐私机制在长期效用上优于非隐私估计。
AI 中文摘要
隐私保护学习通常基于这样一种观点：保护用户数据可维持信任，进而促进用户参与，长期来看能提升效用，但这一主张迄今尚未得到形式化。与此同时，表演式学习为研究“部署后会影响后续观测数据的学习系统”提供了框架。本研究将这两个视角结合，引入“表演式隐私”概念，即数据泄露会减少未来用户参与。我们研究了一个简单模型：智能体（agent）反复贡献数据用于均值估计，但若其数据被泄露则可能退出系统。隐私通过差分隐私（Differential Privacy）机制实现，在估计噪声与未来参与之间形成权衡。通过对该动态的理论研究与数值实验，我们表明：当泄露与参与之间的反馈回路足够强时，有限的隐私预算在长期内的表现优于非隐私估计。这提供了首个证据，证明差分隐私不仅作为保护机制是最优的，从长期效用视角来看亦是如此。
英文摘要
Privacy-preserving learning is often motivated by the idea that protecting users' data can preserve trust and thus participation, improving utility in the long term. However, this claim has not been formalized so far. In parallel, performative learning provides a framework for studying learning systems whose deployment affects the data they later observe. In this work, we bring these two perspectives together and introduce performative privacy, where data leakage reduces future participation. We study a simple model where agents repeatedly contribute data for mean estimation but may leave the system when their data is leaked. Privacy is implemented through differentially private mechanisms, creating a trade-off between estimation noise and future participation. We show, through a theoretical study of the dynamics and numerical experiments, that a finite privacy budget can outperform non-private estimation in the long term when the feedback loop between leakage and participation is sufficiently strong. This provides first evidence that differential privacy can be optimal not only as a protection mechanism, but also from the perspective of long-term utility.
CommentsAccepted at EuroTDP 2026
