It’s a Kind Of Magic: Computation on Ciphertexts
summary
Computing on ciphertext actually produces usable aggregates: a demo uses Paillier-style additive HE so three organizations exchange encrypted rows, a central decryptor returns cohort size 28 and total cost 5004 to derive an average, showing the protocol works end-to-end. Open-source libraries mentioned include Latigo, Zama and OpenFHE; fully homomorphic options are slower and larger. Product architecture uses on-prem data stations with a centralized collaboration service, collaborative key generation and enforced governance rules like minimum cohort size 20 and query rate limits.
AI-written summary · about
chapters (9)
- 0:00Intro
- 2:30Agenda
- 3:101. Computing with ciphertexts
- 11:52Demo
- 14:262. What can you build with it?
- 19:05Questions
- 22:523. What else do you need in practice?
- 31:20Demo
- 39:50Outro
related talks
38mPrivacy in the Open Source: Homomorphic Encryption
01h 07mLattices and Encrypted Computation
40mTrust No One
44mIP Protection and Privacy in LLM: Leveraging Fully Homomorphic Encryption
10mCryptography 10 Years Later: Obfuscation, Proof Systems, and Secure Computation
23mBatchCrypt: Efficient Homomorphic Encryption for Cross-Silo Federated Learning
01h 19mWhen Data Can’t Move but Insight Must
29mProtecting data vs systems: practicality, performance, and problems...