Naive Bayes MNIST Digit Generator

This project demonstrates how a Bernoulli Naive Bayes classifier can be repurposed as a generative model to create synthetic handwritten digits from the MNIST dataset.

📌 Overview

While Naive Bayes is typically used for classification, this project uses it in a generative way:

• Learns the pixel-wise probability of each digit
• Uses sampling to generate new digits
• Produces smoothed, readable digit images

📂 Project Structure

NaiveBayes_MNIST_Generator/ ├── images/ # Generated digit images (0–9) ├── model/ # Saved Bernoulli Naive Bayes model ├── NaiveBayes_MNIST_Generator.ipynb ├── requirements.txt └── README.md

💡 Key Features

• Uses the MNIST dataset
• Trains a Bernoulli Naive Bayes model
• Generates digits using probabilistic sampling
• Outputs both raw and smoothed versions
• Saves model and results

✅ How to Run

1. Clone this repo or download the folder.
2. Install dependencies:
   pip install -r requirements.txt
3. Open NaiveBayes_MNIST_Generator.ipynb
4. Run all cells and check the images/ folder for generated digits.

---

🧠 Inspiration

This was completed as part of the Infosys Springboard course - Generative Models for Developers.

---

👩‍💻 Author

• Swati Upadhyay
• Aspiring AI Engineer | ML Intern @ Trecent Systems