Discrete gradient methods for solving nonconvex, nonsmooth optimisation problems in image analysis

Add to your list(s) Download to your calendar using vCal

• Erlend Skaldehaug Riis (University of Cambridge)
• Wednesday 22 November 2017, 16:00-17:00
• MR14, Centre for Mathematical Sciences.

If you have a question about this talk, please contact Kasia Wyczesany.

We will give an introduction to the discrete gradient method, which is a novel optimisation technique for solving convex and nonconvex variational problems in image processing. Using discretisation tools from geometric numerical integration, these methods are designed to preserve the dissipation of gradient flow systems in a uniformly stable manner. In this talk, we will discuss how discrete gradient methods connects to, and compares to, other methods in gradient-based, as well as derivative-free, optimisation. Particular emphasis will be on nonsmooth, nonconvex optimisation analysis, using the Clarke subdifferential. We will also motivate the results with examples from image analysis.

This talk is part of the Cambridge Analysts' Knowledge Exchange series.

This talk is included in these lists:

• All CMS events
• CMS Events
• Cambridge Analysts' Knowledge Exchange
• DAMTP info aggregator
• Interested Talks
• MR14, Centre for Mathematical Sciences
• My seminars
• bld31

Note that ex-directory lists are not shown.