Lecture 7 - Secure Multiparty Computation for Machine Learning


In this lecture, we look into advanced cryptographic protocols that are used for Secure Multiparty Computation (SMPC). We consider in particular secret sharing, oblivious transfer and garbled circuits. We then see how these protocols can be used to enhance the privacy of machine learning models, especially in the setting of MLaaS (Machine Learning as a Service) and federated learning.

Covered Topics:

  • Basics of Secure Multiparty Computation (SMPC)
  • Card-based game for the 2-party secure AND computation
  • Hard-core predicates and randomized RSA
  • 1-2 Oblivious Transfer from RSA
  • Garbled Circuits
  • Secret Sharing Schemes and their combinatorial characterization
  • SMPC for private Machine Learning
  • Federated Learning

Reading Material

Nice and short textbook on SMPC, freely available on the authors’ websites:

Survey on privacy-preserving machine learning:

Lecture Recordings

The videos are presented below in logical order, although chronologically they have been recorded in different lectures (specifically: introduction to secret sharing at the end of lecture 5, oblivious transfer and garbled circuits in lecture 7, applications to ML in lecture 8).