Speaker
Description
Machine-Learned Likelihood (MLL) is a method that, by combining modern machine-learning classification techniques with likelihood-based inference tests, allows to estimate the experimental sensitivity of high-dimensional data sets. We extend the MLL method by including the exclusion hypothesis tests, and show that the addition of Kernel Density Estimators avoids the need to bin the classifier output in order to extract the resulting one-dimensional signal and background probability density functions. We first test our method on a toy model of multivariate Gaussian distributions, where the true probability distribution functions are known. We then apply it to a case of interest in the search for new physics at the LHC, in which a Z' boson decays into lepton pairs, comparing the performance of MLL for estimating 95% CL exclusion limits to the prospects reported by ATLAS at 14 TeV with a luminosity of 3 ab^{-1}.
