Fully Homomorphic Encryption for Statistical Modeling

Fully homomorphic encryption lets several parties fit statistical models together without revealing their raw records or summaries to each other. This paper introduces two R packages: openfhe.R, which wraps the OpenFHE C++ library (exact integer and approximate real arithmetic, Boolean circuits, multiparty key generation), and homomorpheR, which provides master/worker primitives for multiparty protocols. Any standard R routine whose cross-site quantity decomposes as a sum can then run on encrypted arithmetic. Examples include distributed stratified Cox regression and federated Cox-lasso via consensus ADMM, where only the cross-site update is encrypted, plus federated similarity search and two-party prediction. Parties are assumed honest-but-curious, and threshold key generation means no single party holds the secret key.

Preprint Code Project