KICP seminar: Philipp Frank (Max Planck Institute)

11:00 am–12:00 pm ERC 501

Host: Christoph Welling

Philipp Frank (Max Planck Institute) "Signal reconstruction for fields - a probabilistic perspective"

Many inference tasks in observational astronomy take the form of reconstruction problems where the underlying quantities of interest are fields (functions of space, time and/or frequency) that have to be recovered from noisy and incomplete observational data. These problems are in general ill-posed, as different field configurations may be consistent with the observed data, and therefore their solutions are inherently of probabilistic nature. Approaching them using Bayesian forward modeling by combining an accurate model of the observing instrument and adaptive Gaussian Processes to incorporate prior physical knowledge gives rise to an expressive and flexible framework to solve inverse problems. By making use of Variational methods they can be turned into robust and scalable approximate solvers that have successfully been applied to numerous astrophysical areas such as radio interferometry, Faraday tomography, X- and gamma-ray multi-frequency imaging, Astroparticle physics, and 3D Galactic tomography. I will give a brief overview on the key concepts involved in this approach, and show and discuss selected applications as well as current and future methodology developments.

Event Type


Mar 28