New Local Differentially Private Protocols for Frequency and Mean Estimation

Explore new local differentially private protocols for frequency and mean estimation in this 49-minute lecture by Jelani Nelson from the University of California, Berkeley. Delivered at IPAM's EnCORE Workshop on Computational vs Statistical Gaps in Learning and Optimization, the talk delves into privacy-preserving techniques for distributed applications. Discover how these protocols address challenges in scenarios such as training ML models for autocomplete based on device-stored text history or understanding common app settings across millions of users. Learn about the importance of local differential privacy as the gold standard for ensuring privacy in distributed databases. Examine new protocols developed for two critical problems in this framework: frequency estimation and mean estimation. Gain insights from Nelson's collaborative research with Hilal Asi, Vitaly Feldman, Huy Le Nguyen, and Kunal Talwar, exploring innovative approaches to balancing aggregate data analysis with individual privacy protection in large-scale distributed systems.