This document discusses privacy protection techniques for machine learning models, including differential privacy and encrypted computing. It notes that companies now face large fines for privacy violations. Various types of attacks on privacy are described, including membership inference attacks. TensorFlow provides packages for privacy-preserving machine learning using techniques like federated learning, differential privacy, and encryption. Neural structured learning can improve model robustness against adversarial attacks. The document demonstrates differentially private machine learning using TensorFlow Privacy and compares approaches like Bolton differential privacy that aim to maximize utility while preserving privacy.