Data-Driven Regularisation for Solving Inverse Problems - Carola-Bibiane Schönlieb, Turing/Cambridge

Explore data-driven regularisation techniques for solving inverse imaging problems in this 48-minute talk by Carola-Bibiane Schönlieb from the Alan Turing Institute and University of Cambridge. Delve into the combination of model-based and purely data-driven image processing approaches, starting with "shallow" learning for computing optimal parameters in variational regularisation models through bilevel optimization. Investigate various methods utilizing deep neural networks to tackle inverse imaging problems. Gain insights from the speaker's 2019 Acta Numerica paper, co-authored with Simon Arridge, Peter Maass, and Ozan Öktem. Understand the growing intersection between statistics and computer science in the era of Big Data, and how this cross-fertilization has led to the development of algorithmic paradigms underpinning modern machine learning, including gradient descent methods, generalization guarantees, and implicit regularization strategies.

Data-driven regularisation for solving inverse problems - Carola-Bibiane Schönlieb, Turing/Cambridge