ABSTRACT

A recurrent neural network is a network that performs computational tasks (recognition, association, and so forth) on a given pattern via interaction between a number of interconnected units characterized by simple functions. One class of neural networks which has received a great deal of attention in the literature, and which we have ad­ dressed extensively in the preceding chapters, is described by equa­ tions of the form

x(t) = -C x( t ) + TS(x(t)) + I (7.1.1)

where a; is a real n-vector (which denotes the state variables associ­ ated with the neurons), I is a real n-vector (representing bias terms), C is a real n x n diagonal matrix (representing self-feedback terms), T is a real n x n matrix (representing neuron interconnections), and S(x) is a real n-vector valued function (whose components are sig­ moidal nonlinearities representing the neurons).