This repository utilizes Fully Homomorphic Encryption (FHE) to enable machine learning over encrypted data. In this code we implement a new privacy-preserving solution to training and prediction for trees, proposed by A. Akavia, M. Leibovich, Y. S. Resheff, R. Ron, M. Shahar, and M. Vald [1].
The code uses the CKKS implementation by SEAL, an open source library provided by Microsoft.
- First, update and install c++ tools:
sudo apt-get update
sudo apt-get upgrade
sudo apt install build-essential cmake g++
- Install and build SEAL library (version 4.0.0)
git --branch 4.0.0 https://github.com/Microsoft/SEAL.git
cd SEAL
cmake -S . -B build
cmake --build build
sudo cmake --install build
Make sure that the SEAL installation path matches a path listed in CMakeLists.txt under the find_package() command
To build the project, run the following commands:
cd fhe_random_forest
cmake -S . -B build
cmake --build build
This command will create a 'build' subdirectory and store the executables in it.
After each change to the code, make sure the tests still run successfully:
cd build
ctest -C
After building and passing the tests, you can run the code example:
cd build
./fhe_random_forest
You can also use the provided Dockerfile to build the code with its dependencies on Docker, then run the code example and the tests
cd fhe_random_forest
docker build -t <image name> .
docker run <image name>
docker run <image name> ctest -C
We would like to thank all the code contributors to this project:
Max Leibovich, github: https://github.com/smerte
Omer Sadeh, github: https://github.com/Omer-Sadeh
Boaz Sapir, github: https://github.com/boazsapir
Margarita Vald, email: margarita.vald@cs.tau.ac.il
[1] A. Akavia, M. Leibovich, Y. S. Resheff, R. Ron, M. Shahar, and M. Vald. 2022. Privacy-Preserving Decision Trees Training and Prediction. ACM Trans. Priv. Secur. 25, 3, Article 24 (August 2022), 30 pages. https://doi.org/10.1145/3517197