SecureMNIST

Secure Federated Learning Project using MNIST dataset - B.Tech Data Science

🔐 Secure Federated Learning on MNIST Dataset

A privacy-preserving machine learning project that simulates Federated Learning using the MNIST dataset across multiple virtual clients without sharing raw data.

---

📌 Project Overview

This project demonstrates how federated learning works by training local models on two separate datasets (simulating two clients) and combining their updates using Federated Averaging (FedAvg) — all while ensuring data privacy.

---

🚀 Features

• Federated training using two clients
• Local training with private datasets
• Federated Averaging algorithm
• Accuracy & loss reporting
• Trained model saved as global_model.pth

---

🧰 Tech Stack

• Python 3.9
• PyTorch
• Torchvision
• Visual Studio Code

---

🗂️ Project Structure

secureMNIST/
├── train.py                 # Main training script
├── global_model.pth         # Trained global model
├── data/                    # MNIST dataset
└── README.md                # Project documentation

---

🧪 How It Works

1. Load and split the MNIST dataset into two equal parts.
2. Each client trains a local model on its data.
3. Use FedAvg to average weights and build a global model.
4. Evaluate the global model’s performance.

---

🧑‍💻 Usage

✅ Requirements

pip install torch==1.10.0+cpu torchvision==0.11.1+cpu

▶️ Run the Training

python train.py

📦 Output Example

🔒 Starting Secure Federated MNIST Training...
🏋️ Training on Client 1...
🏋️ Training on Client 2...
🔗 Performing Federated Averaging...
✅ Federated Averaging Done!
🎉 Secure Federated Training Completed.
✅ Accuracy: 72.16%
📉 Average Loss: 1.5850
💾 Global model saved as 'global_model.pth'

---

📸 Screenshots

You can include VS Code training output, data split view, or loss graphs here.

---

🔐 Real-Life Use Case (Healthcare)

Hospitals in different cities want to collaborate on building a diabetes detection model, but cannot share patient data due to privacy laws.

✅ With Federated Learning, each hospital trains a model locally and shares only model updates — not sensitive data. These are averaged to build a global AI model collaboratively.

---

📈 Future Improvements

• Extend to multiple clients
• Add secure aggregation with PySyft or CrypTen
• Visualize training accuracy in real-time
• Apply on medical, finance, or industrial datasets

---

📄 License

This project is licensed under the MIT License – see the LICENSE file for details.

---

Built with ❤️ by Basavavani