Reza Shadmehr, Ph.D.
Professor, Dept of Biomedical Engineering
Johns Hopkins University

A vector calculus for neural computation in the cerebellum:

Neurons in the brain are active not just during execution of a behavior, but also before its onset and after its conclusion. The leading theory that explains the meaning of these activities is that of null spaces. It posits that each neuron contributes to an aspect of behavior with a weight that remains constant in time. The weighted combination of activities across neurons produces a sum that is zero both during the preparation of that behavior and after its conclusion, defining a null space, but non-zero as the behavior is executed, defining a potent space. However, are these weights merely the fitted parameters of an equation, or do they reflect a fundamental aspect of anatomy? Here, we show that in the cerebellum, the weights can be independently estimated for each neuron via spike-triggered methods. These weights define potent and null vectors that not only largely agree with the predictions of the theory, but conform to the principle of superposition. Using these weights to understand the computations in the cerebellum reveals that during a saccade, the activity in the potent space relies on efference copy to predict when the movement is about to reach its target, sending a signal to stop that movement. 
 

Faculty Host: Daeyeol Lee