Direct photon identification with artificial neural network in the photon spectrometer PHOS

A neural network method is developed to separate direct photons from neutral pions in the PHOS spectrometer of the ALICE experiment at the LHC collider. The neural net has been taught to distinguish different classes of events by analyzing the energy profile tensor of a cluster in its eigenvector coordinate system. The Monte-Carlo simulation shows that this method diminishes the probability of π0-meson misidentification as a photon by an order compared with the direct photon detection efficiency in the energy range up to 120 GeV.