This repository contains examples that are used in my dissertation.
Install the requirements to run these scripts by running
pip install -r requirements
In Chap_3/running_example.py
you can find an example of Gaussian Process Regression on the function botorch
's SingleTaskGP
.
This example allows us to show how a GPR returns to the prior away from the support:
In Chap_6_and_7/vae.py
you can find an implementation of a Variational Autoencoder (VAE), trained only on the digits 1 of MNIST
. It illustrates how a VAE lacks proper uncertainty quantification, since the variance shown in latent.jpg
is low away from the support.
Chap_6_and_7/vae_geometric.py
shows how we can modify a VAE to have better uncertainty quantification.