The DIRAC experiment at CERN is designed to measure the life-time of atoms consisting of two oppositely charged pions. Given the corrsponding low production cross-sections and life-time, a high beam intensity and a very efficient trigger are required for a precise measurement. A neural network first level trigger is developed
and used for that purpose. Both the neural network algorithm used and its actual hardware implementation are described. The system uses the fast plastic scintilator information of the DIRAC spectrometer. It is based on a feed-forward network with only 2 nodes in a single hidden layer. This gave the best results compared to more complicated structures specially as far as generalization was concerned. The system is build around custom
made neural network modules developed for earlier applications, proving on the field the flexibility of such algorithms. In 210 ns it selects events with two particles having low relative momentum with efficiency more than 0.94. The corresponding rate reduction for background events is a factor of 2.5. Since its installation in winter 2000 the system is an integral part of the DIRAC trigger scheme. Artificial Intelligence