Machine Learning and Privacy

This lecture covers Federated Machine Learning, focusing on the setup where clients train a global model without sharing their data. It discusses adversarial models, membership and property inference attacks, defenses, and privacy risks. Additionally, it explores Differential Privacy in Machine Learning, including techniques like Objective Perturbation, Output Perturbation, and Gradient Perturbation, as well as the challenges of advanced composition and evaluating differentially private learning.