Generative AI _NVIDIA.pdf# 🧠 Generative AI with Diffusion Models - NVIDIA Assessment (MNIST)
This project was completed as part of the "Generative AI with Diffusion Models" course by NVIDIA. The main objective is to build a diffusion-based generative model capable of creating realistic handwritten digits (0–9) using the MNIST dataset.
Train a generative model that can produce images of handwritten digits, which are recognized with at least 95% accuracy by a pre-trained MNIST classifier (accuracy > 99%).
The model successfully achieved a 96.0% final accuracy, meeting the certification requirement:
✅ NVIDIA certificate earned successfully!
The model uses a conditional U-Net architecture with classifier-free guidance. It learns to denoise images step-by-step through a reverse diffusion process.
- Dataset: MNIST (28x28 grayscale images)
- U-Net Network: Handles time-step and label conditioning
- Diffusion Process: Forward noise addition and reverse denoising
- Conditioning: Digit label is encoded and injected at each step
- Guidance: Uses classifier-free guidance to control generation
🎉 Thanks to NVIDIA for the opportunity to explore cutting-edge generative AI techniques. Proud to have earned the certificate in Generative AI with Diffusion Models! 🧠✨