Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn about Joint Moment Estimation (JME), a novel method for continually and privately estimating both first and second moments of data while maintaining differential privacy guarantees in this Google TechTalk. Discover how JME reduces noise compared to naive approaches by utilizing the matrix mechanism and joint sensitivity analysis, enabling second moment estimation with no additional privacy cost. Explore the theoretical foundations of this approach and understand how it improves accuracy while preserving privacy constraints. Examine two practical applications: estimating running mean and covariance matrices for Gaussian density estimation, and implementing model training with DP-Adam on the CIFAR-10 dataset. Gain insights into advanced differential privacy techniques and their applications in machine learning through this presentation by Nikita Kalinin from the Privacy in ML Seminar series.
