A MULTIINPUT MULTIOUTPUT FUNCTIONAL ARTIFICIAL NEURAL-NETWORK

A generic two-layer feedforward functional neural network is proposed that processes functions rather than point evaluations of functions. S pecifically, the network receives n functions as inputs and delivers m real values as outputs. Its architecture is derived using the nonline ar system identification techniques of Zyla and de Figueiredo. As such , neurons are represented by Volterra functions in Pock space, which i s a reproducing kernel Hilbert space, with synaptic weights that are f unctions themselves. The main advantage is that this functional networ k call be used in the modeling of real-world (continuous-time paramete r) nonlinear systems, capturing the dynamics presented in them, as wel l as in the simulation of their behavior in a computer-based environme nt. (C) 1996 John Wiley and Sons, Inc.