Research

Watch · narrated whiteboard episodesL3

Homomorphic Encryption and Confidential AI

Fully homomorphic encryption computes directly on ciphertext, so a model you do not trust can run inference on a prompt it never sees in the clear — at a real, quantifiable performance tax. This series builds FHE from the noise-growth problem and bootstrapping through the scheme families, private neural-network inference, and an honest comparison with trusted hardware and multi-party computation. Grounded in Gentry, BGV, and CKKS.

Murali Chillakuru·5 episodes
  1. 16 min Episode 1Computing on Ciphertext: The FHE Promise, Noise Growth, and Why Bootstrapping Is the Central IdeaA moderator and a cryptography expert unpack the almost-magical idea of computing on encrypted data — why it's mathematically possible, the noise that threatens every operation, and the bootstrapping breakthrough that made it unlimited.
  2. 17 min Episode 2The Scheme Families: BGV/BFV Exact Integers versus CKKS Approximate RealsA moderator and a cryptography expert compare the main homomorphic encryption schemes — exact integer arithmetic in BGV and BFV, approximate real-number arithmetic in CKKS — and explain why CKKS became the natural fit for private machine learning.
  3. 17 min Episode 3Bootstrapping and Its Cost: Refreshing Noise and the Parameter, Security, Performance TriangleA moderator and a cryptography expert dig into FHE's performance reality — why bootstrapping dominates the cost, the parameter-security-performance triangle every deployment must balance, the levers that cut the bill, and what private inference actually costs today.
  4. 18 min Episode 4Private Inference on Neural Networks: Approximating Non-Linearities Under FHEA moderator and a cryptography expert walk through running a neural network under encryption — why the linear layers are a natural fit, why activation functions are the real obstacle, how polynomial approximation bridges the gap, and what's feasible today.
  5. 18 min Episode 5FHE versus TEEs versus MPC: An Honest Comparison for Confidential LLM ServingA moderator and a cryptography expert close the series by comparing the three approaches to private computation — homomorphic encryption, trusted execution environments, and multi-party computation — their trust models, trade-offs, and how to choose or combine them for private AI.