Jarkko KARI


Jonathan MILLS

James M. NYCE


Gabriela QUEIROZ


Philip WELCH

Brain Dynamics

Abstract. Classical neural nets research has emphasized the structure and the role of the individual neuron, either through trying to infer the formal function that would be best approximated by a given biological neuron, or through using synaptic adaptation as inspiration for automatic learning procedures in computation. This lead to a bottom-up engineering approach, and quite naturally to initial examples of emergent behavior in artificial networks. Such emergence or self-organization is most evident in recurrent networks, which are believed to mimic the structure of the biological brain better than non-recurrent ones. However, the dynamical aspects of those processes involving neural populations have been viewed as somewhat secondary in computation. The dynamics has mainly consisted in periodic cycles of synchronized populations, or in transients to equilibrium states. A modern stream of research engages in considering the so-called dynamical complexity as revealed e.g. by the measurement of the brain's electrical activity. We provide examples of experiments with the biological brain, and review our own work and that of others in trying to assess the computational role of the brain's complex dynamics, as well as obtaining inspiration for possible artificial computing paradigms.