Speaker
Alexander Krull
(University of Birmingham)
Description
Removing imaging noise is an essential problem in scientific applications, where sensors are often pushed to the edge of what is possible. The past years have seen a range of machine learning methods proposed: supervised approaches, using image pairs to learning a mapping from noisy to clean images; Self-supervised approaches, capable of learning such a mapping from noisy data alone; and finally generative approaches capable of additionally capturing the inherent uncertainty of the problem.
In this talk will talk about how these approaches can be understood and derived from a probabilistic perspective.
