The Saddle Point Accountant for Differential Privacy in Machine Learning

Explore the concept of differential privacy in machine learning through this 41-minute Google TechTalk presented by Shahab Asoodeh. Delve into the Saddle Point Accountant method, beginning with an introduction and motivation behind the topic. Gain a clear understanding of key definitions and composition principles. Examine the Saddle Point Approximation technique and its application in privacy accounting. Analyze the accounting algorithm and compare it with other approaches. Engage with thought-provoking questions and conclude with insights into the future of differential privacy in ML applications.