minimal-vae

Minimal PyTorch implementation of a Variational Autoencoder (VAE) trained on MNIST.

Features

• Minimal and readable implementation
• Reparameterization trick
• KL divergence loss
• MNIST training and sampling
• Pure PyTorch implementation
• Educational focus

Repository Structure

minimal-vae/
├── vae.py
├── main.py
└── README.md

Theory and Derivations

Detailed explanations and mathematical derivations are available in the accompanying blog post:

• https://riteshrm.github.io/posts/variational-autoencoder/

The blog covers:

• Variational Inference
• ELBO derivation
• Reparameterization trick
• KL divergence
• Latent space sampling
• VAE training objective

Notes

This repository is intentionally minimal and focuses on understanding Variational Autoencoders rather than large-scale training or state-of-the-art results.