tobias-liaudat/sampling-tutorials

Imaging Inverse Problems and Bayesian Computation - Python tutorials to learn about (accelerated) sampling for uncertainty quantification and other advanced inferences

sampling-tutorials : The Bayesian Approach to Inverse Problems in Imaging Science

In these tutorials, you will learn about Bayesian computation for inverse problems in imaging science. We set up an image deconvolution problem and solve the inverse problem by using different sampling algorithms. Because we obtain samples from the posterior distribution we are able to do uncertainty quantification and other advance inferences. Currently, there is a Python notebook using a vanilla Langevin sampling algorithm (MYULA_pytorch.ipynb) and an accelerated algorithm SK-ROCK (using an explicit stabilized method, SKROCK_pytorch.ipynb). We showcase a deblurring problem using a Total Variation (TV) prior.

Authors

• Dobson, Paul pdobson@ed.ac.uk
• Kemajou, Mbakam Charlesquin cmk2000@hw.ac.uk
• Klatzer, Teresa t.klatzer@sms.ed.ac.uk
• Melidonis, Savvas sm2041@hw.ac.uk

Funding

We acknowledge funding from projects BOLT, BLOOM and LEXCI: This work was supported by the UK Research and Innovation (UKRI) Engineering and Physical Sciences Research Council (EPSRC) grants EP/V006134/1 , EP/V006177/1 and EP/T007346/1, EP/W007673/1 and EP/W007681/1.