| Speaker: Prof. Jari Kaipio Affiliation: University of Auckland Time: 10:00 Wednesday, 16 January, 2019 Location: 303-257 |
| Inverse problems induced by partial differential equations and related boundary value problems are notoriously sensitive to uncertainties in the geometry and the boundary conditions. Furthermore, real world inverse problems practically always necessitate the truncation of the (computational) domain; and on these boundaries, the boundary conditions depend on the material coefficients outside the computational domain. In this talk, we focus on electrical impedance tomography and pose a stochastic nonlocal boundary condition, called the Dirichlet to Neumann map on the truncation boundary. This allows to truncate the computational domain to essentially include the region of interest only. We also consider the problem of unknown boundary geometry and the Bayesian approximation error approach briefly. |