A theory on a neural net with nonmonotone neurons

A theoretical equation on dynamical processes on a neural net consisting of neurons with two-stage nonlinear dynamics is shown. The neurons have nonmonotone response characteristics when parameters are chosen as such. By the exact solution, the high performance of the auto-associative memory net consisting of nonmonotone neurons within a general framework is proved. A correspondence of the dynamical recalling processes of nonmonotone neurons with the learning rule of orthogonal-projection type is shown. Based on the correspondence, it is possible to understand intuitively why nonmonotone neurons are effective