EVENT DETAILS AND ABSTRACT

Title: Neural computation in the cerebellum (and elsewhere): what silent synapses tell us
Speaker: Jean-Pierre Nadal
Speaker Info: ENS, Paris
Brief Description:
Special Note: More current information may be available at Plan-it Purple
Abstract:

The cerebellum is a part of the brain involved in the control of coordinated movements. It has a very specific architecture organized around the Purkinje cells, each one receiving some 150,000 synaptic inputs from granule cells. These granule cell to Purkinje cell synapses are thought to store learned information that enables the cerebellum to perform motor control. Recordings of unitary connections in rat adult cerebellar slices have revealed an unexpected property: a large majority (80%) of these synapses appear to be electrically silent - the responding synaptic weights are zero. In order to investigate the principles governing synaptic weights, we have made use of the (classical) analogy between the Purkinje cell and the perceptron, a prototypical learning machine that learns input-output associations. We have derived the perceptron weights distribution, and compared it with the empirical data. The experimental and theoretical distributions agree closely, and the theoretical analysis reveals a computational role for silent synapses: they are required for maximizing storage capacity. I will present and explain this apparently paradoxical result, and discuss under what conditions silent synapses can be expected to be found in other parts of the brain.