JET ANALYSIS BY NEURAL NETWORKS IN HIGH-ENERGY HADRON-HADRON COLLISIONS

We study the possibility to employ neural networks to simulate jet clustering procedures in high energy hadron-hadron collisions. We concentrate our analysis on the Fermilab Tevatron energy and on the k(perpendicular to) algorithm. We employ both supervised and unsupervised neural networks. In the first case we consider a multilayer feed-forward network trained by the backpropagation algorithm: our results show that these networks can satisfactorily simulate the relevant features of the k(perpendicular to) algorithm. We consider also unsupervised learning, where the neural network autonomously organizes the events in clusters. The results of this analysis are discussed and compared with the supervised approach.