Münster 2017 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 21: Experimentelle Methoden 1 (Computing, Machine Learning, Statistik)
T 21.5: Talk
Monday, March 27, 2017, 17:45–18:00, JUR 253
Development of morphing algorithms for Histfactory using information geometry — •Anjishnu Bandyopadhyay1, Ian Brock1, and Kyle Cranmer2 — 1University of Bonn — 2New York University
Many statistical analyses are based on likelihood fits. In any likelihood fit we try to incorporate all uncertainties, both systematic and statistical. We generally have distributions for the nominal and ± 1 σ variations of a given uncertainty. Using that information, Histfactory morphs the distributions for any arbitrary value of the given uncertainties. In this talk, a new morphing algorithm will be presented, which is based on information geometry. The algorithm uses the information about the difference between various probability distributions. Subsequently, we map this information onto geometrical structures and develop the algorithm on the basis of different geometrical properties. Apart from varying all nuisance parameters together, this algorithm can also probe both small (< 1 σ) and large (> 2 σ) variations. In this talk, it will be also shown how this algorithm can be used for performing interpolation on Monte Carlo distributions of physical variables.