Consideration on pattern-separating function in a generalized random nerve net consisting of two layers

A two-layered random nerve net with feedforward inhibitory connections is characterized by a particular pattern-separating function, which separates input patterns in the sense of overlap rates. The function depends on mean values and variances of various distributions specifying the nerve net and on the firing rate of the input patterns and the second layer. A theory is derived to investigate this pattern-separating function. The effects of the mean values are considered collectively using the derived theory. The influences of the firing rate, both of the input patterns and the second layer, on the function are investigated. As a result, it is shown that the pattern-separating function is largely influenced by the aforementioned mean values, variances, and firing rates. It is also revealed that an excellent nerve net in the pattern-separating function can be obtained when the mean values and variances of distributions are selected properly