Role

Sole Developer & Researcher

Course

EECE 7397 – Advanced Machine Learning

Objective

This project explored foundational concepts in generative modeling through hands-on implementation of VAE and GAN architectures. The goal was to gain practical experience with model design, training mechanics, and evaluation using the Fashion-MNIST dataset.

Technologies

Python, PyTorch, NumPy, Matplotlib

VAE, GAN model construction & training, Fréchet Inception Distance (FID) for quantitative evaluation

My Contributions

• Independently built both VAE and GAN architectures using PyTorch
• Implemented reparameterization in VAE to enable proper backpropagation
• Tuned GAN loss functions for stable training
• Visualized training outputs and assessed quality using FID
• Completed all stages from data preprocessing to model evaluation and presentation

Results

• Successfully implemented VAE and GAN models to generate grayscale images based on the Fashion-MNIST dataset
• VAE produced smooth and continuous image outputs, while GAN generated sharper and more detailed images
• The project provided a hands-on understanding of generative model behavior, highlighting differences in latent space design and training strategies

GitHub Repository

View GitHub Repository