https://github.com/shreyavijaykumar/federated-learning-on-mnist-bridging-accuracy-gaps-across-noisy-and-clean-clients

Federated Learning on MNIST by combining local models from clean and noisy clients with FedAvg to boost overall accuracy
https://github.com/shreyavijaykumar/federated-learning-on-mnist-bridging-accuracy-gaps-across-noisy-and-clean-clients

ai aiml deeplearning fedavg federatedlearning machinelearning mnist privacypreservingai pytorch torch torchvision

Last synced: about 2 months ago
JSON representation

Federated Learning on MNIST by combining local models from clean and noisy clients with FedAvg to boost overall accuracy

• Host: GitHub
• URL: https://github.com/shreyavijaykumar/federated-learning-on-mnist-bridging-accuracy-gaps-across-noisy-and-clean-clients
• Owner: ShreyaVijaykumar
• License: mit
• Created: 2025-08-26T18:18:04.000Z (about 2 months ago)
• Default Branch: main
• Last Pushed: 2025-08-26T18:54:48.000Z (about 2 months ago)
• Last Synced: 2025-08-27T02:42:08.541Z (about 2 months ago)
• Topics: ai, aiml, deeplearning, fedavg, federatedlearning, machinelearning, mnist, privacypreservingai, pytorch, torch, torchvision
• Language: Python
• Homepage:
• Size: 120 KB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Federated Learning on MNIST – bridging accuracy gaps across noisy and clean clients

Federated Learning (FL) enables multiple clients to collaboratively train a model **without sharing raw data**, ensuring privacy and efficiency.  

This project demonstrates **how FL improves model performance for clients with different data qualities** using the MNIST dataset.  

---

## 🚀 Project Overview  

- **Client 1:** Trains on standard MNIST data (clean, high-quality).  

- **Client 2:** Trains on noisy MNIST data (simulating real-world harder cases).  

- **Challenge:** Client 2’s accuracy is lower due to noisy data.  

- **Solution:** Apply **Federated Averaging (FedAvg)** to merge client models and train collaboratively.  

---

## 📊 Results  

| Client | Accuracy Before FL | Accuracy After FL |

|--------|--------------------|-------------------|

| Client 1 (clean MNIST) | ~77.5% | **97.9%** |

| Client 2 (noisy MNIST) | ~72.5% | **96.3%** |

![Sample Input and Output](assets/ai_ip_op.png)

 **Federated Learning successfully bridges the performance gap** between clean and noisy data clients.

 

---

## ✅ Insights

 ![FL accuracy across rounds](assets/ai_graph.png)

 ![Accuracy Comparison between Clients with before and after FL](assets/ai_finalresult.png)

 

---

## 📂 Dataset  

- **MNIST Handwritten Digit Dataset** (60,000 train, 10,000 test)  

- Source: [MNIST on torchvision](https://pytorch.org/vision/stable/generated/torchvision.datasets.MNIST.html)  

- **Why MNIST?**  

  - Classic benchmark for computer vision & FL research.  

  - Small, fast to train, but realistic enough to show domain shift when noise is added.  

---

## 🌍 Real-World Use Cases  

- **Healthcare:** Hospitals with different quality of MRI scans collaboratively train models without sharing sensitive data.  

- **Finance:** Banks collaboratively detect fraud while keeping customer data private.  

- **IoT & Edge Devices:** Smartphones collaboratively improve predictive keyboards without sending personal text data to the cloud.  

---

🧠 **Model Architecture**

- Simple Neural Network (fully connected):

- Input: 28×28 flattened image

- Hidden Layer: 128 neurons (ReLU)

- Output: 10 classes (digits 0–9)

---

📈 **Key Learning Outcomes**

- Understand data heterogeneity problem in Federated Learning.

- Implement Federated Averaging (FedAvg).

- Demonstrate how collaborative training boosts generalization across clients.

- **For clear explanation, read my [blog on medium](https://medium.com/@shreyavj.05/how-federated-learning-solves-the-accuracy-gap-in-heterogeneous-data-8739470447e0)**

---

## 🛠️ Installation & Usage  

###**To install & use this repo **

```bash

# Step-1: to clone the repo

git clone https://github.com/ShreyaVijaykumar/federated-mnist.git

cd federated-mnist

# Step-2: Create Virtual Environment & Install Dependencies

python -m venv venv

source venv/bin/activate  # On Windows: venv\Scripts\activate

pip install -r requirements.txt

# Step-3: Run the Training

python federated_mnist.py