
複数のプライバシー特性を組み合わせて、より強力なプライバシー保証を提供するための新しい枠組みを開発しました。


coremsg

rényi-differential-privacy-group-privacy-amplification-and-unified-amplification-by-subsampling


Rényi Differential Privacy: Group Privacy Amplification and Unified Amplification by Subsampling


title_rewrite


The author exposes the privacy risks associated with topic modeling by demonstrating how membership inference attacks can identify training data members. The approach highlights vulnerabilities in generative models like Latent Dirichlet Allocation and proposes differential privacy solutions.


privacy-risks-in-topic-modeling-membership-inference-attacks-exposed


Privacy Risks in Topic Modeling: Membership Inference Attacks Exposed
