Feedforward neural networks for second-level triggering on calorimeter data

Summary form only given. Two types of feedforward neural networks for the recognition of energy deposition patterns in a calorimeter are considered. The first network is constructed on the basis of explicit linear threshold equations, implementable by simple linear threshold neurons. The second network results from a constructive learning algorithm with neurons performing concealed functions. Monte Carlo events have been used for a hypothetical spaghetti calorimeter at LHC (Large Hadron Collider) and for the Zeus uranium calorimeter at HERA for learning, and in the design and testing of the neural networks. Both algorithms can be realized with a mesh network of digital signal processors, like the Texas Instrument TMS320C40. The construction of a second-level trigger for an LHC experiment based on this processor seems feasible. >