This repository contains scripts, data and a report for VAE based music and image generation, for the course "Techniques Avancées d'Apprentissage", given by Stéphan Clémençon @ENSAE ParisTech in Spring 2018. Note that the repository has to be cloned in order to visualise the python notebook.

This project aims at trying to generate music (and in a first time images) thanks to a particular type of generative model : Variational Auto-Encoders. This method was first described by Diederik P. Kingma and Max Welling in Auto-Encoding Variational Bayes (https://arxiv.org/pdf/1312.6114.pdf) and today achieves state of the art results along with Generative Adversarial Networks (GANs) in data generation (text, image, music). Below are two examples of celebrity images generated with VAEs:

Contacts : charles.dognin@ensae.fr / samuel.ritchie@ensae.fr