Métodos de análise da robustez de redes neurais artificiais sujeitas a retardo no tempo

Abstract

This work presents sufficient conditions for analysis of asymptotic and exponential stability of a class of artificial neural network (ANN) subject to constant or timevarying delays and polytope-bounded uncertainties.The approaches proposed is the type of delay-dependent and the methodology is based on four points: the selection of slack matrices that express the influence of the Newton-Leibniz condition; the appropriate definition of Lyapunov-Krasovskii functionals; the use of linear matrix inequalities (LMIs) and the use of tools of convex optimization to solve problems described in LMI terms. Several examples are presented that corroborate with the theory presented of analysis of the stability of ANN with time-varying delay.