A neural model of timed response learning in the cerebellum

Abstract

A spectral timing model is developed to explain how the cerebellum learns adaptively timed responses during the rabbit's conditioned nictitating membrane response (NMR). The model posits two learning sites that respectively enable conditioned excitation and timed disinhibition of the response. Long-term potentiation of mossy fiber pathways projecting to interpositus nucleus cells allows conditioned excitation of the response's adaptive gain. Long-term depression of parallel fiber-Purkinje cell synapses in the cerebellar cortex allows learning of an adaptively timed reduction in Purkinje cell inhibition of the same nuclear cells. A spectrum of partially timed responses summate to generate an accurately timed population response. In agreement with physiological data, the model Purkinje cell activity decreases in the interval following the onset of the conditioned stimulus, and nuclear cell responses match conditioned response (CR) topography. The model reproduces key behavioral features of the NMR. including the properties that CR peak amplitude occurs at the unconditioned stimulus (US) onset: a discrete CR peak shift occurs with a change in interstimulus interval (ISI) between conditioned stimulus (CS) and US; mixed training at two different ISIs produces a double-peaked CR; CR acquisition and rate of responding depend unimodally on the ISI; CR onset latency decreases during training; and maladaptively timed, small-amplitude CRs result from ablation of cerebellar cortex.