A neural particle discriminator based on a modified ART architecture

A compact neural discriminator using a modified ART architecture is presented, envisaging the online classification of particles in a high event rate collider experiment. For making the online operation feasible, the neural discriminator is fed with projected data onto segmented principal components, reducing deeply the dimensionality of data input space. The designed discriminator is able to extract the physics of interest in the experiment (detecting high energy electrons with an efficiency higher than 99.4%) from the huge background noise generated by the collider machine, with a false alarm probability smaller than 2%. The discriminator system is being implemented on digital signal processor technology, achieving a processing speed of 68.8 μs.