A Digital Neuromorphic Implementation of Cerebellar Associative Learning

Abstract

The cerebellum is a neuronal structure comprising half the neurons of the central nervous system. It is essential in motor learning and classical conditioning. Here we present a digital electronic module, pluggable to an artificial autonomous system, designed following the neural structure of the cerebellum. It emulates the associative learning function as described in the context of classical conditioning. Building on our previous work we propose a neuromorphic implementation portable to a Field Programmable Gate Array (FPGA), capable of generating responses of variable amplitude. To validate our design we test it with the simulation of a robot performing a navigation task on a curvy track. Our digital cerebellum is able to make adaptively-timed rotations with variable amplitude suitable for the track. This suggests that the Purkinje cell dependent learning circuits of the cerebellum do not only time the triggering of actions but can also tune the specific response amplitude.