A 2D implementation of RealNVP (Real-valued Non-Volume Preserving), a normalizing flow model that learns invertible transformations between complex data distributions and simple latent distributions, enabling density estimation and sampling.
This repository contains two Jupyter notebooks:
real_nvp_2d.ipynb
Includes:
- RealNVP model construction and training
- forward and inverse transformations
- density estimation and outlier detection
- sampling and data generation
real_nvp_2d_interactive_demo.ipynb
Interactively move a point in the data space and observe how it is mapped to the latent space by the trained RealNVP model.
Run in browser:
