A discrete approach for a model of temporal learning by the cerebellum: in silico classical conditioning of the eyeblink reflex.

Cerebellar cortex is known to be involved in acquisition and expression of eyeblink conditioned reflex. These phenomena imply temporal intervals of learning. Several cellular and network mechanisms have been proposed to produce the eyeblink. In this paper we briefly review the main theories concerning temporal coding, and we propose an alternative way of producing and storing delays and signal sequences after supervised learning. A network of Leaky Integrate-and-Fire (LIF) neurons is built, taking into account several cerebellar features. This network is then trained to produce simple or multiple eyeblink delays using (i) the classical conditioning paradigm and (ii) known data on cerebellar spike timing dependent plasticity (STDP). The resulting model behaves like an adaptive temporal filter. It improves cell subpopulations effects according to their mean firing rate. This rate based selection allows robust supervised learning of temporal events (i.e., delayed signals) and gives the network ability to react with anticipation on the arousal of a noxious event.