homomorphic encryption
In plain wordsA way of locking data so that a computer can still work with it while it stays locked: the computer never sees the real numbers, and only the owner can open the answer.
Encryption whose algebraic structure survives computation: operating on ciphertexts yields an encryption of the corresponding result on the plaintexts, so data can be processed by a party that cannot read it. Partially homomorphic schemes support one operation, such as addition (Paillier) or multiplication (unpadded RSA); fully homomorphic encryption (FHE) supports both, and hence arbitrary circuits, by periodically 'bootstrapping' to control the noise that lattice-based ciphertexts accumulate. Schemes such as BGV, BFV, TFHE and CKKS, the last designed for approximate real-number arithmetic, make encrypted machine-learning inference possible: a client sends an encrypted input and receives an encrypted prediction that the server cannot read. Costs are steep — often thousands of times slower than plaintext, with non-linear functions such as ReLU or softmax approximated by polynomials — so practical uses are small models, private lookups and components of larger protocols.
The server ran the classifier under homomorphic encryption and never saw the patient's data or the diagnosis.
Origin
Ronald Rivest, Leonard Adleman and Michael Dertouzos proposed 'privacy homomorphisms' in 1978. Craig Gentry constructed the first fully homomorphic scheme in his 2009 Stanford dissertation and STOC paper. Microsoft's CryptoNets (2016) demonstrated neural-network inference on encrypted data.
Related
- secure multi-party computationsecure multi-party computation, n.: Cryptographic methods that let several parties compute a result from their combined private data — say, an average salary — without any of them revealing its own data to the others.
- confidential computingconfidential computing, n.: Running computations inside a hardware-sealed compartment of a chip, so that data and AI models stay protected even from the cloud company or administrator operating the machine.
- privacy-preserving machine learningprivacy-preserving machine learning, n.: The family of techniques for training and using AI models while keeping the underlying personal or confidential data hidden — from adding careful noise to computing on encrypted data.
- differential privacydifferential privacy, n.: A mathematical promise that the results of analyzing a dataset — or a model trained on it — would look almost the same whether or not any one person's data was included, so little can be learned about that person.
- confidential computingconfidential computing, n.: Running computations inside a hardware-sealed compartment of a chip, so that data and AI models stay protected even from the cloud company or administrator operating the machine.
Sources
- Gentry, C. (2009). Fully Homomorphic Encryption Using Ideal Lattices. ACM STOC 2009.
- Rivest, R. L., Adleman, L. & Dertouzos, M. L. (1978). On Data Banks and Privacy Homomorphisms. In Foundations of Secure Computation. Academic Press.
Mentioned in
confidential computingconfidential computing, n.: Running computations inside a hardware-sealed compartment of a chip, so that data and AI models stay protected even from the cloud company or administrator operating the machine., privacy-preserving machine learningprivacy-preserving machine learning, n.: The family of techniques for training and using AI models while keeping the underlying personal or confidential data hidden — from adding careful noise to computing on encrypted data., secure multi-party computationsecure multi-party computation, n.: Cryptographic methods that let several parties compute a result from their combined private data — say, an average salary — without any of them revealing its own data to the others.