Bayesian Inference in Collider Physics
by
The Standard Model is currently under intense scrutiny at the LHC as we search for New Physics. Machine learning techniques have been widely adopted to aid in this pursuit. We examine in this talk why Bayesian Inference techniques can offer unique advantages for analyzing HEP data. We analyze the use of Bayesian Inference to better understand observables and improve searches at the LHC and other colliders. We will discuss important statistical tools, such as Graphical Models, to aid in implementation.
Compared to many Neural Network and supervised frameworks, Bayesian Inference can considerably reduce the impact of potential biases from Monte Carlo simulations. One of its achievements is the ability to learn from data and correct imprecisions in the simulations. Moreover, Bayesian Inference can extract signal and background fractions and shapes solely from the signal region, eliminating the need for a control region and avoiding potential extrapolation errors. In this talk, we will discuss hypotheses, assumptions and implementation techniques on data for these techniques to be successful. We will present recent results using these techniques in top-tagging, quark/gluon jet tagging, four-top searches, and di-Higgs searches
