Neural networks with stochastic preprocessing for particle recognition in cosmic ray experiments

Neural networks can be used with good results in particle recognition. An application on 4 GeV CERN experimental data of e− and π−, taken with the prototype of the silicon tungsten calorimeter of the WIZARD collaboration, is shown. The full detector is being used for cosmic ray antimatter research. A stochastic preprocessing is used in connection with a neural network (back propagation algorithm). This preprocessing consists in giving as input to the net the probability, for a given discriminating parameter value, to belong to a given particle class. In this way the input layer is normalized and the back propagation algorithm can exploit the relations between different parameters. This results in an increased convergence speed and recognition capability of the net. With this algorithm a high hadron (π−) recognition efficiency (93.6% on experimental data, 97.1% on Monte Carlo data) and a low electromagnetic contamination (0.1% on experimental data and, 3 × 10−4 on Monte Carlo) is reached.