PHYSTAT 2003 was devoted to statistical problems in particle physics, astrophysics and cosmology. Since that meeting, the sophistication of statistical practice in these fields has increased significantly. However, from time to time, it is helpful to put aside our sophisticated tools and remind ourselves of a few core ideas. It is also helpful, occasionally, to look over the shoulders of colleagues in related fields; we may be inspired to approach a problem differently. In this lecture, I begin with the simple, but highly instructive, ON/OFF problem, in order to highlight key differences between the frequentist and Bayesian viewpoints. I then consider a model of image reconstruction (that is, image unfolding) commonly used in astronomy. This will naturally lead to a discussion of the difficult problem of multi-dimensional priors. Next, I discuss the fitting of cosmological models to Type Ia supernovae data which illustrates the phenomenon of model non-identifiability and why, therefore, it is useful to have a way to quantify the effective dimensionality of a parameter space. In view of the growing interest in machine learning in astronomy, If time permits, I end with a few general remarks about machine learning models and algorithms.
The seminar will be done remote only, using ZOOM for this event, link to join the seminar: https://cern.zoom.us/j/222861107
M. Girone, M. Elsing, L. Moneta, M. Pierini
Event co-organised with the PHYSTAT Committee