IFT6135: Representation Learning, third assignment.
This is the third assignment of the IFT6135: Representation-Learning, taught by Prof. Aaron Courville.
Variational autoencoders (VAEs, Questions 1-3), autoregressive models (Question 4), and generative adversarial networks (GANs, Questions 5-7) : Statement and Solution.
Generative models: Statement.
This is a joint work with Abderrahim Khalifa, Yann Bouteiller and Amine Bellahsen.
Problem 1: implementing an estimator for the Jensen-Shannon divergence and another for the Wasserstein distance between two distributions
Solution.
Problem 2: training a Variational Auto-Encoder (VAE) on the binarized MNIST dataset
Solution.
Problem 3: comparing two generative models, the WGAN-GP model and the VAE model using the SVHN dataset
Solution.