Fully Homomorphic Encryption
Fully homomorphic encryption (FHE), first constructed by Gentry in 2009, supports evaluating circuits of arbitrary depth on ciphertexts, with the decrypted result equal to the circuit output on the plaintext.
Purpose
It enables privacy-preserving outsourcing of arbitrary computation, the "holy grail" of cryptography, underpinning encrypted machine learning and encrypted database queries.
Mechanism
It builds on ideal-lattice somewhat homomorphic encryption plus bootstrapping to control noise. Bootstrapping homomorphically evaluates the decryption circuit under an encrypted decryption key, resetting the noise. Later schemes (BFV, CKKS, TFHE) optimize performance.
Use cases
- Encrypted machine-learning inference
- Encrypted SQL queries over databases
- Privacy-preserving statistical analysis