Skip to main content Skip to secondary navigation

Research Overview

Main content start

Using the cerebellar and vestibular circuitry to understand the logic of neural learning algorithms

The computations performed by our neural circuits in support of sensation, action, and cognition are powerfully shaped by the patterns of synaptic connections between neurons.  This synaptic connectivity is not static, but is continuously changing.  The patterns of activity in a neural circuit, elicited by our experiences, can trigger a strengthening or weakening of specific synapses within the circuit.  This synaptic plasticity underlies our ability to learn from experience. We are analyzing the neural learning rules governing the local “decisions” each synapse in a circuit makes on a moment-by-moment basis about whether to alter its strength, based on its pattern of input.  Ultimately, our goal is to understand how these local decisions are coordinated throughout a neural circuit to yield an algorithm for the adaptive regulation of the circuit’s function.

Our research leverages the simplicity of the circuit architecture in a brain region called the cerebellum, which makes systematic analysis of the function of this circuit experimentally and analytically tractable. We have developed a battery of behavioral paradigms in mice for studying the vestibular control of behavior and its adaptive regulation by the cerebellum, including learned changes in the amplitude and timing of movements, the generalization of learning, and factors influencing the persistence of memory. We are addressing these issues by combining our expertise in analyzing neural circuits using electrophysiological, behavioral, and computational approaches with powerful molecular-genetic tools for precisely manipulating specific neurons or synapses in vivo.

An understanding of the algorithms that neural circuits use to tune their own performance as they compute can guide the design of machines with computing and learning capacity more closely approximating humans, and will enable us to optimize learning in health and in neural circuits damaged by injury or disease.