Frequently Asked Questions about FHE
10 months ago
- #privacy
- #cryptography
- #homomorphic-encryption
- FHE allows operations on encrypted data without decryption, ensuring privacy.
- FHE schemes can implement operations like addition, multiplication, and even comparisons through polynomial approximations.
- FHE performance is improving, with applications like facial recognition possible in seconds on a GPU.
- FHE is secure against quantum computers, using similar hardness assumptions as NIST-standardized PQC methods.
- FHE is used in production for specific applications, with more in development.
- FHE does not require encrypting everything; interactions between plaintext and ciphertext data are possible.
- FHE is not suitable for training LLMs on encrypted data but can convert trained models to work on encrypted data.
- FHE's killer app is yet to be discovered, with potential in legal, insider risk, and new service scenarios.
- FHE is compared to local computation, TEEs, and other privacy-enhancing technologies, with trade-offs in security and performance.
- FHE's mathematical structure does not necessarily weaken security, with no successful attacks discovered yet.