🥇 Winner of the 2025 First Place Award

Abstract

Guardin is an application that allows healthcare professionals to easily contribute their data and compute into federate learning projects, allowing models to train on their data while maintaining absolute data security. The application provides a dashboard for the user to find a model they like, and allows them to pull an image of that model from a Docker Registry. Then, the user can drag and drop their data into a box, which automatically mounts that data into the Docker Container. Lastly, once the far server begins training, it will send training messages to the client, upon which the client will start training, and send its weight updates to the server. When the model converges, all clients will be sent a copy of the final weights, which the user can then download and use to reconstruct the final FL model.

Guardin Quick Start

cd guard-in
npm i
npm start

Note that this only starts the client application, training does not start until the server starts it.

For details on how to run the Guardin backend stack yourself, see the Deployment README.

Usage

Guardin provides a user-friendly desktop application for healthcare professionals to participate in federated learning projects while maintaining complete data privacy. The application operates through three main screens:

1. Model Selection

When you launch Guardin, you'll see the Model Selection Screen featuring available federated learning models:

Each model displays:

• Description: What the model does and its intended use case
• Data Types: Required input data format (Tabulated Data, Genomic Data, etc.)
• Supported Formats: File formats accepted (CSV, VCF, FASTA, etc.)
• Use Cases: Medical specialties and applications (Diagnostics, Oncology, etc.)

To select a model: Click on any model card to proceed to data upload.

2. Data Upload

The Data Upload Screen allows you to securely provide your dataset:

• Drag & Drop: Simply drag your data file into the upload area
• File Browse: Click "Browse Files" to select from your file system
• Format Validation: Guardin automatically validates your data and mounts it to the container
• Privacy Protection: Your data never leaves your local environment

Data Security: Your original data files remain on your local machine and are only temporarily mounted into secure LOCAL Docker containers for processing.

3. Federated Learning Training

The Testing Screen shows real-time federated learning progress:

Real-time Monitoring

• Live Output Stream: View real-time training logs and progress updates
• Training Metrics: Monitor accuracy and loss values as they improve over rounds
• Visual Graphs: Interactive charts showing model performance over time
• Round Progress: Track which training round the federated learning is currently on
• Elapsed Time: See how long the training process has been running

Training Process

1. Initialization: Your local client connects to the federated learning server
2. Local Training: The model trains on your local data within the secure Docker container
3. Weight Updates: Only model weights (not your data) are shared with the federation
4. Aggregation: The server combines weights from all participating clients
5. Model Updates: Your client receives the updated global model for the next round
6. Convergence: Process repeats until the model reaches optimal performance

Results and Download

• Model Weights: Download the final trained model weights when training completes
• Performance Metrics: View final accuracy, loss, precision, recall, and F1-score
• Training Summary: Get a complete overview of the training session
• ZIP Download: Receive model weights packaged as a convenient ZIP file

Key Features

🔒 Data Privacy: Your sensitive healthcare data never leaves your local environment. Only encrypted model weights are shared.

🐳 Docker Integration: Automatic Docker container management ensures consistent, isolated training environments.

📊 Real-time Visualization: Interactive graphs and metrics provide immediate feedback on training progress.

🌐 Federated Learning: Participate in collaborative AI training with institutions worldwide while maintaining data sovereignty.

⬇️ Model Export: Download trained model weights to deploy in your own healthcare systems and applications.

Workflow Example

1. Launch Guardin and browse available federated learning models
2. Select "Alzheimer's Predictor" for early diagnosis research
3. Upload your CSV file containing de-identified patient data
4. Monitor training progress as your data contributes to the global model
5. Download final weights when federated learning completes
6. Deploy the model in your clinical decision support systems

This workflow enables healthcare institutions to collaboratively improve AI models while ensuring patient data remains completely private and secure.

Contribute

Contributions are welcome! If you'd like to contribute, please open an issue or submit a pull request. See the contribution guidelines for more information.

Support

If you have any issues or need help, please open an issue or contact the project maintainers.

License

This project is licensed under the MIT License.