NICE (Non-linear Independent Components Estimation) is a flow-based generative model for density estimation and sample generation, leveraging reversible transformations for efficient computation.
This repository implements both Additive and Affine coupling layers, tested on MNIST and Fashion-MNIST datasets. Results include loss plots and generated samples.
For more details, refer to the original paper NICE: Non-linear Independent Components Estimation By Laurent Dinh, David Krueger, and Yoshua Bengio (2015) Link