Utility-Optimized Local Differential Privacy Mechanisms for Distribution Estimation

Explore a 21-minute conference talk from USENIX Security '19 that introduces Utility-Optimized Local Differential Privacy (ULDP) mechanisms for distribution estimation. Delve into the innovative approach presented by Takao Murakami from AIST, which aims to enhance the utility of personal data statistics while maintaining privacy protection. Learn about the limitations of traditional Local Differential Privacy (LDP) and how ULDP addresses these issues by providing equivalent privacy guarantees only for sensitive data. Discover two proposed ULDP mechanisms: utility-optimized randomized response and utility-optimized RAPPOR. Examine the concept of personalized ULDP with semantic tags, designed to accommodate varying sensitivity levels among users. Gain insights into the theoretical and experimental evidence demonstrating the superior utility of ULDP mechanisms compared to existing LDP approaches, especially when dealing with a high proportion of non-sensitive data. Understand how ULDP can achieve near non-private mechanism utility in low privacy regimes when most data is non-sensitive.