Storage capacity and errors of neural networks

An estimate of the storage capacity of a neural network working with real images is obtained for neural networks based on correlation optics. It takes into account not only the number of neurons in the network, but also the structure of patterns in the form of the correlation radius of their images. Methods from the theory of statistical solutions and estimation theory are used to determine the errors of the first and second kinds, as well as the error in determining the coordinate of a pattern. The statistical variability of the input pattern, the correlation radius of the image, and the number of neurons and patterns in the memory are taken into account. (C) 2000 The Optical Society of America. [S1070-9762(00)00502-9].