homomorpheR: privacy-preserving statistics across sites

homomorpheR runs statistical computations across sites that never share their data, using fully homomorphic encryption through the openfhe.R interface to OpenFHE: CKKS for real-valued arithmetic, BFV and BGV for exact integers, with n-of-n threshold key generation so that no single party can decrypt.

Details

The protocol actors are a Master and its Sites. A LocalSite, built with make_worker(), holds its data and a contribution_fn; master_aggregate() runs one round, in which each site returns its contribution already encrypted and only the aggregate is decrypted. Build the master with make_ckks_master() when one party may hold the secret key and with make_threshold_master() when none may. Ordinary R modeling code – stats4::mle(), stratified survival::coxph(), convex programs via CVXR – then runs unchanged with the encrypted round as its objective. The package vignettes develop each protocol in full.

A frozen implementation of the Paillier additive scheme is kept for backward compatibility; see paillier_keypair().

See also

Useful links:

• https://bnaras.github.io/homomorpheR/

• Report bugs at https://github.com/bnaras/homomorpheR/issues

Author

Maintainer: Balasubramanian Narasimhan naras@stat.Stanford.EDU [copyright holder]

Authors:

• Balasubramanian Narasimhan naras@stat.Stanford.EDU [copyright holder]

Examples

## A Poisson rate estimated across three sites: each site encrypts
## its negative log-likelihood, and only the sum is decrypted.
local_nll <- function(data, lambda)
    -sum(stats::dpois(data, lambda, log = TRUE))
y <- c(9, 12, 7, 11, 10, 8, 13, 9, 10, 12)

cc   <- openfhe.R::fhe_context("CKKS", multiplicative_depth = 1L,
                               scaling_mod_size = 50L, batch_size = 8L)
keys <- openfhe.R::key_gen(cc)
master <- make_ckks_master("Master", crypto_context = cc, keypair = keys)
set_workers(master, list(
    make_worker("S1", y[1:3],  local_nll),
    make_worker("S2", y[4:6],  local_nll),
    make_worker("S3", y[7:10], local_nll)))

fit <- stats4::mle(function(lambda) master_aggregate(master, lambda),
                   start = list(lambda = 5))
c(encrypted = stats4::coef(fit)[["lambda"]], cleartext = mean(y))
#> encrypted cleartext 
#>      10.1      10.1