Deconstructing VAEs from Scratch

Explore the mathematics of variational inference, latent spaces, and the reparameterization trick — built entirely from scratch in pure PyTorch.

The Math of Latent Spaces

Breaking down the ELBO derivation, KL divergence, variational inference, and why you can't just backpropagate through sampling.

Building Encoder, Decoder, reparameterization, and ELBO loss with both dense and convolutional variants — entirely from scratch.
View Code on GitHub

Visualizing latent space structure, interpolating between digits, and comparing reconstruction quality across latent dimensions.