Global asymptotic stable control of growth dynamics of feed-forward neural network

A design of artificial neural networks by using growth dynamics is proposed. The growth dynamics exist when a hidden neuron is added to a network, and the connection strength is treated as a control input. The objective of controlling the growth dynamics is to find a controller that achieves a global asymptotic stable (GBS) property, i.e. the zero error convergence. Theoretical foundation of the growth control including the existence of GAS controllers is developed. A practical method that guarantees to produce GAS controllers is also proposed. A function approximation problem is used as an example to demonstrate the feasibility of this approach