Segmented self-organized feature extraction for online filtering in a high event rate detector

In this work, a novel feature extraction strategy is proposed for the electron/jet channel of ATLAS detector second level trigger. Placed around one of the collision points of LHC (the next generation particle accelerator), ATLAS will be responsible for the selection and recording of relevant information. A huge amount of data will be generated, in spite of that, only a few will be relevant for characterization of new physics. Considering this, an efficient triggering system is very important to maximize the detector performance. A segmented signal processing routine is proposed here in order to make use of different characteristics of each detector layer. Self-Organizing Maps (SOM) were trained for each layer, and further adjusted through Learning Vector Quantization (LVQ) to maximize particle discrimination. Neural classifiers perform electron/jet identification using as inputs the Segmented SOM information. Through the proposed approach, a discrimination efficiency of 97.4% was achieved for a false alarm probability of 2.4%.