Neural network with switching units and its application in HEP (INVITED)

Abstract

We introduce a separation algorithm of a neural network with switching units (NNSU) and present its applicability to data mining applications. NNSU algorithm exploits principles of artificial multilayer perceptron like neural networks and linear decision trees. Combination of both approaches provides more flexible tools for set separation. The training phase of NNSU uses a genetic optimization of NNSU architecture, which is in general an acyclic oriented graph, hence a convenient definition of corresponding genetic fitness function allow us to tune separation results to meet user defined requirements. We tested usability of NNSU model in different problems in high energy physics data processing, among others Higgs boson search. In the next we discuss shortly the possibility of efficient hardware implementation. Owning to the fact that response phase of NNSU algorithm behaves as a piecewise linear discriminant the testing phase of the algorithm can be performed by hardware means only. First step toward hardware implementation was done, and possibility of full electronic implementation was studied. On the base of this study we estimate that event processing speed about 25 mega-samples per second is reachable by full hardware implementation without significant degradation of separation quality.