This repository implements the Variational Recurrent Neural Network (VRNN) in PyTorch. There is a demo with the MNIST dataset.

This implementation is from A Recurrent Latent Variable Model for Sequential Data, Junyoung Chung et. all (https://arxiv.org/abs/1506.02216). The following image from the paper illustrates model's operations and its objective function.

Requirements

• The repository was created with Python 3.11 but newer versions compatible with the other libraries should also work.
• PyTorch 2.9.1
• pytorch-lightning 2.5.6
• torchmetrics 1.8.2
• torchvision 0.24.1

Feel free to open issues and discussions for improvements.