Local independent component analysis applied to highly segmented detectors

A novel particle discrimination strategy is proposed in this work for the ATLAS detector High-Level Trigger. The available data set, composed by electron and jet signatures, was clustered using Self-Organizing Maps and Local Independent Components were estimated for each group. A hybrid neural-genetic structure was used as classifier. Considered performance improvement was achieved with the proposed approach, 97.5% of electrons were correctly identified for 3 % jet misclassification.