I will begin by overviewing the Bayesian approach
to the reconstruction of fields from indirect
and noisy (possibly nonlinear) measurement functionals [1].
I will then explain the basic Bayesian level set approach to reconstructing piecewise constant fields [2].
Finally I will demonstrate how the method can be enhanced
by means of a hierarchical multiscale approach in which the
charateristic length scale of interface separation is
learned from the data, along with the geometry of interfaces themselves.
[1] M. Dashti and A.M. Stuart. The Bayesian approach to inverse problems.
To appear in Handbook of Uncertainty Quantification, Springer, 2016.
http://arxiv.org/abs/1302.6989
[2] M.A. Iglesias, Y. Lu, A.M. Stuart, "A level-set approach to Bayesian
geometric inverse problems", submitted.
http://arxiv.org/abs/1504.00313
Joint work with Matt Dunlop (Warwick) and Marco Iglesias (Nottingham)