No label defined (Q4246006)

The authors obtain three approximation results of functional analysis. Let \(\sigma\) be the bounded generalized sigmoid function and denote by \(C(M)\) the set of all continuous mappings defined on \(M\). The typical result of this paper is embodied in the following Theorem: If \(K_1\subset X\), \(K_2\subset\mathbb{R}^n\), \(U\subset C(K_1)\) are compact sets and \(T: U\to C(K_2)\) is a nonlinear continuous operator, where \(X\) is a Banach space, then for every \(\varepsilon> 0\) there exist natural numbers \(M\), \(N\), \(m\), real numbers \(\theta_{ik}\), \(c_{ik}\), \(\xi_{ijk}\), \(x_j\in K_1\), \(i= 1,\dots, M\), \(j= 1,\dots, m\), \(k= 1,\dots, N\), such that \[ \Biggl|T(u)(y)- \sum^N_{k=1} \sum^M_{i=1} c_{ik}\sigma \Biggl(\sum^m_{j=1} \xi_{ijk} u(x_j)+ \theta_{ik}\Biggr)\Biggr|< \varepsilon,\quad \forall u\in U,\;y\in K_2. \] The authors claim that the approximation results obtained provide a theoretical tool for the applications of neural networks in dynamic system modeling.