LMS Workshop: Variational Methods Meet Machine Learning

Theme

Only recently have machine learning algorithms came into focus in the inverse problems and more general computational mathematics communities. The early publications include both attempts to provide theoretical underpinning of the methods (e.g. analysis of deep network frameworks both more general as well as through construction of dedicated "interpretable" layers, or analysis of stochastic gradient methods), as well as proposals of how the methods could be used in the context of inverse problems (e.g. unsupervised learning of parameters of variational models or training deep neural networks to provide approximate solutions of variational problems). The goal of this one day workshop is to bring together researchers working on variational methods in inverse problems and machine learning to identify the parallels between the fields as well as discover new research avenues. This will be facilitated by two discussion sessions.

Invited Speakers

• Thomas Pock (TU Graz)
• Andreas Hauptmann (University College London)
• Mila Nikolova (Centre de Mathématiques et de Leurs Applications - CNRS - ENS Cachan)
• Michael Hintermüller (Humboldt University of Berlin)
• Joana Grah (University of Cambridge)
• Jonas Adler (KTH Royal Institute of Technology)
• Christian Etmann (University of Bremen)

Schedule

Please click on the titles for abstracts where available.

10:00 - 10:30	Registration
10:25 - 10:30	Welcome/Opening Remarks
10:30 - 11:00	Thomas Pock	Learning better models for inverse problems in imaging
11:05 - 11:35	Andreas Hauptmann	Model based learning for accelerated, limited-view 3D photoacoustic tomography
11:35 - 12:05	Coffee break
12:05 - 12:35	Jonas Adler	Learned iterative reconstruction schemes, theory and practice
12:40 - 13:10	Christian Etmann	Regularisation of neural networks with input saliencies
13:15 - 13:30	Morning discussion
13:30 - 14:30	Lunch
14:30 - 15:00	Michael Hintermüller	Bilevel optimization and some "parameter learning" applications in image processing
15:05 - 15:35	Joana Grah	Learning filter functions in regularisers by minimising quotients
15:35 - 16:05	Coffee break
16:05 - 16:35	Mila Nikolova	Fast solvers for approximating inconsistent systems of linear inequalities
16:40 - 17:00	Afternoon discussion

Registration

To register for the workshop please email Marta Betcke: m.betcke[at]ucl.ac.uk. There is no fee, but we would like to know if you are coming for planning purposes. Also, if you would like to participate in the poster session, which will take place during lunch, please include a proposed title and abstract for your poster in the email.