Learning about variational autoencoders by applying them to a synthetic dataset – a 2D spiral with isotropic additive noise. A writeup related to this codebase can be found on my website.

Relevant papers:

• The original paper by Kinga and Welling
• The longer introduction by the same authors
• Taming VAEs for more details on training

To get started with the code

git clone https://github.com/mckuzyk/vae_demo.git
cd vae_demo
uv run scripts/basic_run.py