The non-linear signal domain selection using a new quality function in neural net training

Applying cuts in the space of event parameters is the traditional technique for background rejection in high energy physics. Obtained by considering the simulation of signal and background events, particular values of these cuts are used to reach required balance between efficiency of signal detection and background rejection. experiments in high energy accelerator physics and astrophysics are operating with multidimensional parametric spaces. Thus, the problem of the best cut selection is of vital interest. Frequently used rectangular cuts are usually too restrictive and can deteriorate the shape of selected multivariate signal distribution. With the aid of the proposed method, it is possible to obtain smooth nonlinear shape of signal cluster which optimizes the ratio of signal to noise. The search of the best γ-cluster on the data files of Crab nebula detection by Atmospheric Cherenkov Telescope of Whipple collaboration proves the superiority of neural techniques upon traditional methods.