yibinzlh / manifold_mixture_vaes

Manifold Learning by Mixture Models of VAEs for Inverse Problems

Manifold Learning by Mixture Models of VAEs for Inverse Problems

This repository contains the code for the paper "Manifold Learning by Mixture Models of VAEs for Inverse Problems" [1] available at
http://arxiv.org/abs/2303.15244
Please cite the paper, if you use the code.

This repository contains code for approximating the charts of a manifold by a mixture model of VAEs. More precisely, the following examples are implemented.

• The script manifold_vae_toy.py, manifold_vae_bar.py and manifold_vae_balls.py contain the source code for training the mixtures of VAEs in Section 5, 6.1 and 6.2.

• The script move_on_toy.py reproduces the trajectories from Figure 6.

• The scripts move_on_bar_fig.py and move_on_bar_trajectories.py reproduce the deblurring experiments from Figure 7 and 8.

• The script move_on_balls_calderon.py reproduces the EIT experiment from Section 6.2.

The code is written with PyTorch 1.12.0. The EIT experiment uses version 2019.1.0 of the Fenics library.

For questions, bugs or any other comments, please contact Johannes Hertrich (j.hertrich(at)math.tu-berlin.de).

Citation

[1] G.S. Alberti, J. Hertrich, M. Santacesaria and S. Sciutto.
Manifold Learning by Mixture Models of VAEs for Inverse Problems.
Arxiv preprint 2303.15244.