Usecase

This is a Rust implementation of a usecase for the Elisabeth stream cipher. this runs homomorphic inferences on transciphered data from the F-MNIST dataset.

Prerequisite

To run this project, you will need the Rust compiler, and the FFTW library. The compiler can be installed on linux and osx with the following command:

curl  --tlsv1.2 -sSf https://sh.rustup.rs | sh

Other rust installation methods are available on the rust website.

To install the FFTW library on MacOS, one could use the Homebrew package manager. To install Homebrew, you can do the following:

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install.sh)"

And then use it to install FFTW:

brew install fftw

To install FFTW on a debian-based distribution, you can use the following command:

sudo apt-get update && sudo apt-get install -y libfftw3-dev

You can then clone this repository by doing:

git clone git@github.com:princess-elisabeth/elisabeth_usecase.git

Moreover, this project requires access to the closed beta of Zama's Virtual Machine. To gain access to this beta, please contact hello@zama.ai and ask to be added to both zamavm-beta and concrete_internal-beta repositories.

Usage

Before running any test or benchmark, you should export the following RUSTFLAGS:

export RUSTFLAGS="-C target-cpu=native"

Then run the project with the following command:

cargo run --release -- *NUMBER_OF_INFERENCES*

Where NUMBER_OF_INFERENCES should be replaced by the actual number of inferences you want to run.

Note that an inference should be several minutes long.