Training of the dynamic neural networks via constrained optimisation

The paper deals with training of a dynamic neural network by using an algorithm which takes into account constraints on network parameters. A neural network considered is composed of dynamic neurons, which contain inner feedbacks. To train this network, a stochastic approximation method is applied. The stability analysis during training is also investigated. As a result of this analysis, a learning algorithm based on a constrained optimization has been elaborated. Efficiency of a proposed approach is presented using an example of modelling of an unknown non-linear dynamic system.