Using neural networks to learn energy corrections in hadronic calorimeters

Abstract

I present two methods to determine the energy correction factors for the ATLAS hadronic calorimeter. In the first we use a recurrent neural network with nearest neighbour feedback in the input layer and, in the second, an event classification step by a competitive network precedes the learning of the correction factor. A comparison with a normal feed-forward net with backpropagation learning scheme is presented for the first method.