Federated Learning & Blockchain for Vehicular Event Recognition

Collaborative Traffic Event Recognition Experiments — Dockerized Environment

This repository contains the full implementation of the experiments described in the article:
“Collaborative Traffic Event Recognition with Federated Learning: A Blockchain-Oriented Framework.”

It includes:

• Centralized training code (baseline)
• Federated Learning experiments using Flower (FLwr)
• A Dockerized environment with GPU support
• Scripts to run, manage, and connect to containers
• Configuration files for running multi-client FL experiments

---

✅ Repository Structure

├── centralised/ # Centralized training experiments
│ ├── entrenamiento.py # Main training script
│ └── ...
├── fl-blockchain/ # Flower Federated Learning project
│ ├── pyproject.toml # FL client/server config & experiment parameters
│ ├── ...
├── Dockerfile # GPU-enabled Docker image
├── run.sh # Script to launch a new Docker container
├── connect.sh # Script to connect to an existing container
└── README.md

---

🚀 1. Building the Docker Image

This environment uses a GPU-enabled Docker configuration.

To build the Docker image, run:

docker build --build-arg USERNAME=$(whoami)-docker \
             --build-arg USER_UID=$(id -u) \
             -t <image_name>:latest .

Replace:

• <image_name> → name of your image (e.g., flblockchain)

---

🚀 2. Launching the Docker Environment

Use the provided script:

./run.sh <gpu_device> <container_name> <image_name>

Examples:

Use a specific GPU

./run.sh 4090 mycontainer flblockchain
# In our setup we have both NVIDIA RTX 4090 and NVIDIA TITAN so you should adjust the `run.sh` file to meet your set up 

Use all GPUs (recommended for servers with 2+ GPUs)

./run.sh all mycontainer flblockchain

---

🔗 3. Connecting to an Existing Container

If the container is already running:

./connect.sh <container_name>

Example:

./connect.sh mycontainer

---

⚙️ 4. Flower (FLwr) Environment Setup

After building the Docker image, Flower requires an update to your ~/.bashrc.

Inside the running container, run:

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc
source ~/.bashrc

This ensures Flower commands and Python local binaries work correctly.

---

🧪 5. Running Centralized Experiments

The centralized baseline code lives in the centralised directory.

Move to that folder:

cd centralised

Then run the training:

python3 entrenamiento.py

This will execute the centralized CNN model used as a baseline in the article.

---

🌐 6. Running Federated Learning Experiments (Flower + Docker)

The fl-blockchain directory contains the complete Flower project.

Navigate to the folder:

cd fl-blockchain

To run the FL experiments, follow Flower’s official deployment guide: 👉 Flower Documentation

Important notes:

• The number of clients, rounds, FL parameters, and experiment variables are set in:

fl-blockchain/pyproject.toml

• Changing the number of clients in pyproject.toml must match the number of worker services you deploy using Flower.
• You may launch clients and server in separate terminals inside the Docker environment (or separate containers, depending on your setup).

---

📌 7. Requirements

All dependencies are already baked into the Docker image.

Key frameworks include:

• Flower (FLwr)
• TensorFlow 2.19
• Keras
• Docker + NVIDIA Container Toolkit
• Python 3.10
• GPU acceleration (CUDA)

---

🤝 Contributions

Contributions, pull requests, and issue reports are welcome. Please feel free to extend or adapt the FL experiments.

---

🤝 Acknowledgements

This work was supported by:

• 🇪🇸 Spanish Research Agency (AEI), Project Serverless4HPC (PID2023-152804OB-I00)
• 🇪🇺 Generalitat Valenciana, CIACIF/2022/176, co-funded by European Social Fund Plus (ESF+)

---

📬 Contact

For technical issues or collaboration proposals, this are the authors contact information:

Tamai Ramírez-Gordillo 📧 tamai.ramirez@ua.es

Francisco A. Pujol 📧 fpujol@ua.es

Adriana Morejón 📧 adriana.morejon@ua.es

Higinio Mora 📧 hmora@ua.es

🏫 University of Alicante
🌐 https://www.cloudlab.ua.es