MKFL: practical tool for Secure Federated Learning

MKFL is a privacy-preserving federated learning tool based on the xMK-CKKS encryp tion scheme.

• Client data security is provided by the xMK-CKKS homomorphic multi-key encryption scheme [1]. The parameters communicated between the clients and the server are encrypted using the aggregated public key and can only be decrypted by the collaboration of all clients. <!---- Contribution evaluation [2] is realized by a trusted server computing the Shapley value [3] of each client update. We implemented three client elimination methodologies based on it :
  • delete_one: the client with the Shapley value the most under the threshold is discarded.
  • delete : all clients with Shapley values under the threshold are discarded.
  • set_aside : all client with Shapley values under the threshold are discarded but are given the opportunity rejoin the process at the next contribution evaluation. --->

Table of Contents

• Installation
• Usage
• Docker
• Examples

Installation

Works with Python 3.10.12. <!---1. Clone the repository: bash git clone git@github.com:marielonfils/CCAFLR.git 2. Clone the dependencies repositories and install them: bash git clone git@github.com:marielonfils/flower.git cd flower pip install . git clone git@github.com:marielonfils/TenSEAL.git cd TenSEAL pip install. --->

1. Install the dependencies repositories and install them:

Download flower from https://anonymous.4open.science/r/flower-4103/README.md. bash cd flower pip install . Download TenSEAL from https://anonymous.4open.science/r/TenSEAL-8F2A/README.md. bash cd TenSEAL pip install .

1. Install dependencies: bash cd CCAFLR pip install -r requirements.txt pip install pyg_lib torch_geometric torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.4.0+cpu.html pip install torch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1 --index-url https://download.pytorch.org/whl/cpu

Usage

Scripts

The CCAFLR/src/FL folder contains bash scripts to run easily the federated learning task on a single machine: - fl_run.sh : without encryption and without contribution evaluation - flrunenc.sh : with encryption but without contribution evaluation

Do not forget to replace the address of the server at the end of the src/FL/flclientenc.py or src/FL/fl_client.py. <!---- ce_run.sh : without encryption but with contribution evaluation - cerunenc.sh : with encryption and with contribution evaluation. --->

Manually

If you want to run each component manually, you need to run the aggregation server, the contribution evaluation server and the clients.

To run the secure server, use the following command in CCAFLR/src/: python python FL/fl_server_enc.py --nclients [number of clients] --nrounds [number of rounds] --filepath [folder path to store results] --dataset [split_scdg1] [--noce] <!-----methodo [deleteone/delete/setaside] --threshold [threshold value] The noce option specifies that no contribution evaluation should be done. In that case, it is unnecessary to specify the methodo and threshold options.--->

To run the federated learning task without encryption, replace the filenames of the server and the clients by fl_server.py and fl_client.py, respectively. <!--- and remove the enc* option from the CE server command.--->

Do not forget to replace the address of the server at the end of the src/FL/flclientenc.py or src/FL/fl_client.py.

Modify the database used

In the init_datasets function of src/FL/main_utils: - add a branch with the name of the dataset - create a Dataset object and provide it an initialization function - call init_db on the new Dataset and return it

Modify the model used

In the get_model function of src/FL/main_utils: - add a branch with the name of the model - initialize your model - create a Model object and provide your model to it - return the Model object

Docker Installation and Usage

Docker image can be created with bash docker build ./ -t ccaflr Note that the build takes about 25-30 minutes.

To run the secure federated learning framework (without contribution evaluation) on a single machine docker compose can be used bash docker compose up -d

.env file contains parameters for docker compose: number of clients, number of rounds, dataset name, model name, path to the dataset, and path to a folder to output results. Latest execution logs can be checked with bash docker compose logs

Examples

Example datasets are provided in src/databases.

The first one is the directory examples_samy. It contains graphical representation of malware. The dataset to give as parameter is example_samy and the model is GINE.

The second dataset is example_images.zip. It contains image representations of malware. It should first be unzipped in a folder of the same name. The dataset to give as parameter is example_images and the model is images.

References

[1] Ma, Jing, et al. "Privacy-preserving Federated Learning based on Multi-key Homomorphic Encryption." arXiv preprint arXiv:2104.06824 (2021). [2] Liu, Zelei, et al. "CAreFL: Enhancing smart healthcare with Contribution‐Aware Federated Learning." AI Magazine 44.1 (2023): 4-15. [3] Shapley, Lloyd S. "A value for n-person games." (1953): 307-317.