Recent progress in the development of computational systems simulating nervous activity in animals is reviewed. The biological basis of a zonal model of cortical functions is reassessed, with a view to the simulation of unsupervised but motivated learning of motor activity in the cerebellum. A macrozone of the cerebellum is represented as an extended network of unit circuits, each consisting of granule cells and a Purkinje cell, together with several inhibitory interneurons. The potential of each neuron is defined by a set of activation levels, including several refractory and potentiated, as well as resting and firing levels. Learning is achieved by a stochastic process involving the progressive modification of weights associated with the synapses of cells left in potentiated levels by activation. Results obtained from the operation of a computer program (listed as an Appendix to the paper) are described, suggesting, among other things, an event sequencing role for extracellular potential waves generated elsewhere in the brain, and an inverse correlation between speed and accuracy in learning.

HOME