Bilevel Learning for Imaging Problems

Many imaging problems, such as denoising or inpainting, can be expressed as variational regularisation problems. These are optimisation problems for which many suitable algorithms exist. However, the problem of selecting a suitable regulariser is not so straightforward. In this talk, I will introduce new deterministic and stochastic algorithms for bilevel optimisation, and show how they can be used to learn suitable regularisers from training examples.