RustFL: Secure and Asynchronous Federated Learning with Differential Privacy and Secure Multiparty Computation

Project Overview

RustFL is a federated learning framework designed to securely and asynchronously train machine learning models while preserving the privacy of client data. By integrating Differential Privacy (DP) and Secure Multiparty Computation (SMPC), the framework ensures that sensitive information remains confidential throughout the training and aggregation process.

Key Features

1. Privacy-Preserving: Differential Privacy is applied to each model update to obfuscate sensitive information from individual clients.
2. Secure Aggregation: The server performs model updates using encrypted shares, ensuring that no client data is exposed during the aggregation process.
3. Asynchronous Communication: The framework utilizes asynchronous communication between the clients and the server.

Architecture

RustFL operates with a central server and multiple clients, where the clients perform local model training and then update the global model securely and privately.

1. Client Training and Model Update Generation

   Clients retrieve the latest global model from the server.

   Perform local training using their datasets.

   Generate shares of the updated model weights using a Secret Sharing Mechanism (Shamir’s Secret Sharing).

   Apply Differential Privacy to the shares to ensure that individual client data is not leaked.

2. Encryption and Transmission

   After applying differential privacy, the noisy shares are encrypted to ensure further privacy protection.

   The encrypted shares are sent to the server for aggregation.

3. Server Aggregation

   The server receives the encrypted noisy shares from the clients.

   Aggregates these shares using Secure Multiparty Computation (SMPC) techniques.

   Reconstructs the updated global model from the aggregated shares.

   Updates the global model with the new aggregated weights.

Technologies Used

Rust: The main programming language for implementing the federated learning framework.

tokio: Asynchronous runtime for handling concurrent tasks efficiently.

tch (PyTorch bindings): Used to implement deep learning models and tensor operations in Rust.

reqwest: For making HTTP requests between clients and servers.

log: To log important information, warnings, and errors during the process.

Requirements

1. AMD architecture required

2. libtorch file of version 2.2.0 is required(Can doenload from here :https://download.pytorch.org/libtorch/cpu/libtorch-macos-x86_64-2.2.0.zip)

3. pytorch of version 2.2.0 version is required

4. for OpenSSL error: follow the steps

                     Download openssl-3.4.0.tar.gz from github and extract it
                     In the terminal, follow:
                     cd /Absolute/path/to/openssl-3.4.0
                     ./config --prefix=$HOME/openssl --openssldir=$HOME/openssl
                     make
                     make install
                     export OPENSSL_DIR=$HOME/openssl
                     export OPENSSL_LIB_DIR=$OPENSSL_DIR/lib
                     export OPENSSL_INCLUDE_DIR=$OPENSSL_DIR/include
                     export PKG_CONFIG_PATH=$OPENSSL_LIB_DIR/pkgconfig
                     source ~/.bashrc
                     ls $OPENSSL_LIB_DIR
                     ls $OPENSSL_INCLUDE_DIR
                     cd /Absolute/path/to/RustFL
                     cargo clean
                     cargo build
   

5. For torch not found error, follow following in terminal:

                    python3 --version #verify the version and update
                    python3 -m pip install --upgrade pip
                    pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu117
                    source /path/to/venv/bin/activate
                    export LIBTORCH_USE_PYTORCH=1
   

6. For .dyld file not found error:

                  export DYLD_LIBRARY_PATH=/Absolute/path/to/libtorch/lib
   

Running the System

1. Build the Project:

                       cargo build
   

2. Run the Example server:

                       cargo run --bin example_server
   

3. Run the Example Client:

                       cargo run --bin example client
   

4. For Documentation:

                    cargo doc --open
   

5. To build fuzz:

                       cargo fuzz build
   

6. To Run fuzzing:

                       cargo fuzz run fuzz_target_1
   

Logging:

You can enable detailed logging with:

                       RUST_LOG=info cargo run --bin bin_name

We have also developed an application which uses our crate: https://github.com/Sharvani1291/RustFL/blob/main/Example/README.md

Our crate can be downloaded from crates.io: https://crates.io/crates/RustFL

Docker:

To run the example codes in docker container, follow the steps:

To build an image:

                    docker build -t <inage_name> .

To run the container for server:

                    docker run -d --name <container_name_1> -p 8081:8081 <image_name>

To run the container for client:

                    docker run --name <container_name_2> --network="host" <image_name>

Final Release Details

This is the final release of RustFL, where the framework has been fully implemented and optimized. It incorporates Differential Privacy for added privacy protection and Secure Multiparty Computation (SMPC) to ensure that model updates from clients are aggregated securely, maintaining the confidentiality of each client’s data.

The system is designed to be privacy-preserving, offering secure federated learning for decentralized machine learning applications.

Conclusion

RustFL is a secure and efficient federated learning system that provides end-to-end privacy protection through Differential Privacy and Secure Multiparty Computation. By using a decentralized model training approach, this system ensures that sensitive client data remains private while enabling collaborative machine learning.