A spiking neural architecture that learns tasks

Abstract

Cognitive architectures based on neural networks typically use the Basal Ganglia to model sequential behavior. A challenge for such models is to explain how the Basal Ganglia can learn to do new tasks relatively quickly. Here we present a model in which task-specific procedural knowledge is stored in a separate memory, and is executed by general procedures in the Basal Ganglia. In other words, learning happens elsewhere. The implementation discussed here is implemented in the Nengo cognitive architecture, but based on the principles of the PRIMs architecture. As a demonstration we model data from a mind-wandering experiment.