Multidimensional measurements using multivariate techniques - the future of particle physics precision measurements?

Name: Multidimensional measurements using multivariate techniques - the future of particle physics precision measurements?
Start: 2022-12-15T11:15:00+01:00
End: 2022-12-15T12:15:00+01:00
Location: No location set

by Dag Gillberg (Carleton U.)

Thursday 15 Dec 2022, 11:15 → 12:15 Europe/Stockholm

Description

One of the established and robust methods to study properties of a fundamental process is to measure differential cross section that presents the production rate as a function of different kinematic variables. Such distributions need to be “unfolded”, meaning corrected for detector effects, and can then be used to test various hypotheses, for example predictions provided by different MC generators or predictions that include new physics, such as dark matter or anomalous CP asymmetry. The LHC has provided a wealth of such measurements, and the established procedure is to make these results publicly available in the HepData database, with an associated Rivet routine that facilitates the comparison with theoretical predictions. There are however several shortcomings with such measurements: one needs to settle on the exact a) observables, and b) bin boundaries used in the measurements, and they are c) normally performed in 1D, i.e. one observable at the time. (2D or even 3D measurements are occasionally performed, for example measuring the momentum spectrum separately in bins of rapidity.)
This seminar will discuss recent developments in data analysis that attempts to overcome these limitations, by performing unbinned measurements in high dimensional space. Such measurements are also unfolded and come with full uncertainties. They would hence easily reproduce traditional binned measurements, and it would be trivial to also change the binning or do an unbinned analysis. Further, one would be able to construct new observables event-by-event by combining the individual observables. I will discuss the proposed methodology that makes this possible, which makes significant use of recent developments in machine learning. I will discuss associated challenges and shortcomings and implications for making such measurements public.