Training recurrent pulsed networks by genetic and Taboo methods

In this paper, we simulate small recurrent pulsed neural networks (up to a dozen of neurons) of leaky integrate-and-fire (LI&F) neurons and we train them thanks to general optimisation methods: genetic algorithm and taboo search; in a way inspired by the training of artificial neural networks (ANN). Unlike the taboo search, the genetic method succeeds in our training procedure. Yet, it proves out to be unsuitable for mimicking the behaviour of a whole network, which involves all nodes and not only input and output neurons.