A semi-free weighting matrices approach for neutral-type delayed neural networks (Q1003996): Difference between revisions

The paper addresses stability issues of constant delayed neural networks described by neutral delay differential equations in of the form, \[ \dot{x}(t)=-Dx(t)+Af(x(t))+Bf(x(t-\tau))+C\dot{x}(t-\tau)+J, \] with initial condition \(x(t)=\phi(t)\), for \(t \in [-\tau,0]\), and where \(D=\text{diag}(d_i) >0\), is a \(n\times n\) matrix. \(A\), \(B\), \(C\) are \(n\times n\) interconnection matrices representing the weight coefficients of the neurons, \(f=(f_1,\ldots,f_n)^T\) the activation functions and \(J=(J_1,\ldots, J_n)^T\) are external constant inputs. Assuming that the activation functions are bounded and Lipschitz, the main result of the paper is that this delay neural network is globally exponential stable provided that exist certain functions and matrices with precise properties. This result is proof combining several techniques such as semi-free weighting matrices approach, the Lyapunov-Krasovskii functional and a linear matrix inequality. As a corollary of the theorem, semi-free weighting matrices are proposed to use in the system synthesis of the problem, instead of the traditionally free weighting matrices. In this way the number of variational matrices is reduced, the construction of the Lyapunov functional is more convenient and the computational cost is reduced. The paper ends with a couple of examples where, using fewer variables, better stability properties of the method are shown when compared to previous results.