Enhanced Label-Only Membership Inference Attacks with Fewer Queries
Hao Li
34th USENIX Security Symposium (USENIX Security '25) · Day 3 · ML and AI Privacy 2
This talk, presented by Hao Li at USENIX Security, introduces a novel approach to **Label-Only Membership Inference Attacks (MIA)**, significantly reducing the number of queries required while enhancing attack performance. The research, titled "Enhanced Label-Only Membership Inference Attacks with Fewer Queries," addresses critical vulnerabilities in machine learning models that can lead to severe user privacy infringements. Traditional membership inference attacks often demand an impractical number of queries, limiting their real-world applicability and their utility as a tool for security researchers.
AI review
Legitimate ML privacy research with a clean technical contribution — the fixed-BD insight is real and the query efficiency gains are meaningful. But this is a USENIX Security paper talk, not a security practitioner session, and the delta from prior MIA work is incremental rather than paradigm-shifting.