The numerical solution of linear ordinary differential equations by feedforward neural networks (Q1334712): Difference between revisions

The possibility to construct directly and noniteratively a feedforward neural network to approximate arbitrary linear ordinary differential equations is demonstrated. A function approximation theory, which deals with the problem of approximating or interpolating a continuous, multivariate function is considered to explain in general supervised learning and artificial neural network. Transfer functions and Lagrange polynomial splines are considered as basis functions. Input weights and basis weights are introduced. Transformating the transfer function of the neural networks into splines is considered. The method of weighted residuals is described. The method uses the hard limit transfer function. It is linear in storage and processing time, and the \(L_ 2\) norm of the network approximation error decreases quadratically with the increasing number of hidden layer neurons. Two examples are considered. The first one is a first-order linear ordinary differential equation. The second one is a second-order linear differential equation. The result presented uses the hard limit transfer function.