Privacy-Preserving Machine Learning

This lecture covers techniques such as Additive and Shamir Secret Sharing for protecting secrets and enabling computations without revealing sensitive data. It explores operations on secret shared data, including secure multiparty computation protocols. The lecture delves into the combination of Homomorphic Encryption with secret sharing schemes, addressing the challenges of communication-intensive computations. It presents real-world applications in healthcare and discusses the envisioned nation-wide deployment of privacy-preserving techniques. The lecture concludes with a detailed overview of POSEIDON, a system for Privacy-Preserving Federated Neural Network Learning, highlighting its building blocks, evaluation, scalability, and the challenges addressed by the solution.