Automated selection of particle-jet features for data analysis in High Energy Physics experiments
Di Luca A., Follega F.M., Cristoforetti M., Iuppa R.
VI - Fisica applicata, acceleratori e beni culturali
GSSI Ex ISEF - Aula A - Venerdì 27 h 15:30 - 19:00
» Download abstract
Modern high energy physics experiments measure many features of particle jets and a large number of parameters directly enter Physics analyses, where experimenters use a limited fraction of these parameters to select signal against background. This strategy, yet simple and powerful in a cut-and-count approach, maybe be sub-optimal when machine learning strategies are envisaged. Brute force cannot be used neither: training networks with all available jet features is often impossible, due to lack of calibration or estimation of systematic uncertainties. Selecting features to use is still an open problem. We show here that it is possible to rank the relative importance of all available jet features in an automated fashion by engineering a fast and powerful jet classification model. Features are sorted with the Random Forest algorithm, then selected as input quantities for a Deep Learning Neural Network. We make the relation between Random Forest importance ranking and signal-to-background ratio increase explicit, varying the number of features to feed the Neural Network with. We benchmark our procedure with the case of highly boosted di-jet resonances decaying to two $b$ quarks, to be selected against an overwhelming QCD background. Results from Monte Carlo simulation with HEP pseudo-detectors are shown.