Applications of neural networks in high-energy physics

The use of Artificial Neural Networks as a tool in High Energy Physics is investigated. Neural networks are used to analyze data resulting from hadron-hadron collisions. Colliders data contain many observables which can be very cumbersome to analyze if the traditional methods are applied. We present two case studies by applying the neural networks to the production of Higgs bosons and the Top quarks. In both cases, neural networks provide further enhancement in the identification of the signal processes over the background processes. In the investigation of the Higgs boson, neural networks are used to help distinguish the {dollar}ZZ to ellsp+ellsp-{dollar}-jet-jet signal produced by the decay of a 400 GeV Higgs boson at a proton-proton colider energy of 15 TeV from the "ordinary" QCD Z+jets background. The ideal case where only one event at a time enters the detector (no pile-up) and the case of multiple interactions per beam crossing (pile-up) are examined. In both cases, when used in conjunction with the standard cuts, neural networks provide an additional signal to background enhancement. In addition, we investigated the event signature of the {dollar}ell nu bbar b qbar q{dollar} decay mode of the Top-pair production in proton-antiproton collisions at 1.8 TeV. Neural networks and Fisher discriminants are used in conjunction with modified Fox-Wolfram "shape" variables to help distinguish the Top-pair signal from the W+jets and bb+jets background. Instead of requiring at least four jets in the event, we find that it is faster and better to simply cut on the number of calorimeter cells with transverse energy greater than some minimum. By combining these cell cuts with the event shape information, we are able to obtain a signal to background ratio of around 9 while keeping 30% of the signal. This corresponds to a signal to background enhancement of around 370.