A model of cerebellar metaplasticity

Abstract

The term 'learning rule' in neural network theory usually refers to a rule for the plasticity of a given synapse, whereas metaplasticity involves a 'metalearning algorithm' describing higher level control mechanisms for apportioning plasticity across a population of synapses. We propose here that the cerebellar cortex may use metaplasticity, and we demonstrate this by introducing the Cerebellar Adaptive Rate Learning (CARL) algorithm that concentrates learning on those Purkinje cell synapses whose adaptation is most relevant to learning an overall pattern. Our results show that this biologically plausible metalearning algorithm not only improves significantly the learning capability of the cerebellum but is very robust. Finally, we identify several putative neurochemicals that could be involved in a cascade of events leading to adaptive learning rates in Purkinje cell synapses.