Research

Watch · narrated whiteboard episodesL3

Secure Multi-Party Computation for Collaborative AI

Zero-knowledge proves a computation and homomorphic encryption computes on one party's ciphertext; secure multi-party computation lets several distrusting parties jointly compute over their combined private data without any one seeing another's inputs. It is the missing third leg of confidential AI. This series builds MPC from secret sharing and garbled circuits through private set intersection to federated learning with secure aggregation. Grounded in Shamir, Yao, and GMW.

Murali Chillakuru·5 episodes
  1. 19 min Episode 1The MPC Guarantee, Defined: Computing Without Revealing Inputs, and the Adversary ModelsA moderator and a cryptography expert define secure multi-party computation from its foundation — the ideal-world definition that says what it means to compute jointly while learning nothing but the result, the adversary models, and why it matters for collaborative AI.
  2. 19 min Episode 2Secret Sharing and Threshold Cryptography: Shamir, Additive Shares, and Threshold DecryptionA moderator and a cryptography expert build the foundational MPC primitive — Shamir secret sharing that splits a secret into meaningless pieces, additive shares you can compute on, and threshold cryptography that removes single points of failure.
  3. 19 min Episode 3Garbled Circuits and GMW: Two Routes to General Multi-Party Computation and Their Trade-offsA moderator and a cryptography expert explore two classic MPC protocols beyond secret sharing — Yao's garbled circuits that encrypt a whole computation, the GMW protocol's gate-by-gate approach, and the rounds-versus-bandwidth trade-off that decides which to use.
  4. 20 min Episode 4Private Set Intersection and Federated Primitives: The Workhorses of Privacy-Preserving JoinsA moderator and a cryptography expert examine one of MPC's most practical tasks — computing the overlap between two datasets without either party revealing its full contents, how it works, its variants, and the leakage you must reason about.
  5. 20 min Episode 5MPC for Machine Learning: Federated Learning, Secure Aggregation, and the Confidential-Computing TrioA moderator and a cryptography expert close the MPC series by applying it to machine learning — federated learning and secure aggregation, private training and inference, and an honest comparison of MPC against homomorphic encryption and trusted hardware.