Neural online filtering based on preprocessed calorimeter data

Aiming at coping with LHC high event rate, the ATLAS collaboration has been designing a sophisticated three-level online triggering system. A significant number of interesting events decays into electrons, which have to be identified from a huge background noise. This work proposes a highly-efficient L2 electron / jet discrimination algorithm based on artificial neural processing fed from preprocessed calorimeter information. The feature extraction part of the proposed system provides a ring structure for data description. Energy normalization is later applied to the rings, making the proposed system usable for a broad energy spectrum. Envisaging data compaction, Principal Component Analysis and Principal Component of Discrimination are compared in terms of both compaction rates and classification efficiency. For the pattern recognition section, a fully-connected feedforward artificial neural network was employed. The proposed algorithm was able to achieve an electron detection efficiency of 96% for a false alarm of 7%.