Generative models

Variational Auto Encoder

Implementing Variational Auto Encoder (VAE) from the scratch. Use Gaussian and Bernoulli priors to train the model on MNIST dataset and to generate new images. Compare VAE with deterministic Auto Encoder and use Gaussian Mixture Models to transform the latter into a generative model.

Monte Carlo Markov Chains

Implement various sampling methods. Use PyMC to sample from observed data and estimate posterior distribution.