wangz10 / Generative-Models

Tutorial on deep generative models with experiments on MNIST

Lecture on Generative Models

Getting started

1. Run the Jupyter Notebook Server

The Jupyter notebook server can be started through the Docker image jupyter/tensorflow-notebook by running the shell script:

$ ./run_docker.sh

2. Train the Deep Generative Models

GPU is not required to train these model because of the small dataset size and relatively small neural networks. Training of the models can be done by running the Train_VAE_on_MNIST.ipynb and Train_BiGAN_on_MNIST.ipynb notebooks.

3. Run the Main Tutorial

References

• Ng AY & Jordan MI: On Discriminative vs. Generative classifiers
• Kingma & Welling: Auto-Encoding Variational Bayes
• Goodfellow IJ et al: Generative Adversarial Networks
• Donahue et al: Adversarial Feature Learning
• Flow-based Deep Generative Models
• Glow: Better Reversible Generative Models