Modular neural networks for on-line event classification in high energy physics

Abstract We discuss the application of Modular Neural Networks (MNNs) for high-performance, high rate classification of HEP events, both in terms of the algorithms involved, and their hardware implementation. Three different problems were treated successfully with the MNN framework, namely the identification of electrons and photons in the first and second trigger levels of the CMS experiment and the classification of Cherenkov rings in a RICH detector, showing the versatility and conceptual simplicity of MNN for triggering in HEP experiments. A prototype of the electron/photon trigger primitives generation system for the CMS experiment, based on a MNN and implemented with L-Neuro 2.3 chips, was developed and tested in the CERN SPS H4 beam line. The system reached excellent performance, identifying electrons with full efficiency at the 40 MHz LHC clock. The same hardware, properly reprogrammed, is able to handle the trigger of strangelet rings in the context of an experiment to search for exotic matter.