Privacy Technologies

Computing on data you can't see

1:00

How to get answers from data without exposing the people inside it — from re-identification attacks and differential privacy to homomorphic encryption, MPC, zero-knowledge proofs, and federated learning.

By the end, you can

Explain why removing names alone does not anonymize a dataset and define quasi-identifier.
State the differential privacy guarantee in terms of databases D and D′ and describe what epsilon controls.
Describe how the Laplace mechanism adds noise and why accuracy is preserved at the aggregate level.
Explain what homomorphic encryption enables, how bootstrapping extends it to unlimited operations, and why the server never learns the plaintext.
Describe Yao's millionaires problem and how MPC solves it without a trusted third party.
State the three properties of a zero-knowledge proof and explain how zk-SNARKs power zk-rollups.
Describe the federated learning architecture, explain the gradient leakage risk, and name mitigations.
Articulate the privacy-vs-utility trade-off that runs through all these techniques.

Up next in Information Theory, Cryptography & Security