Training artificial neural networks with dynamic particle swarm optimisation (Q2925076)

Artificial neural networks are efficiently trained by use of a particle swarm algorithm mostly with an additional assumption on the statistics of classification problems. The assumption is not correct for many practical applications. Here, a dynamical variant of the particle swarm method is used for training of the artificial neural networks on dynamical errors of the inverse propagation. The efficiency of application of the method proposed in the dynamical classification problem is demonstrated.