The operation of the brain is not 'clockwork', but rather probabilistic. The project will provide proof of concept data for a new theoretical and experimental framework that utilizes this feature, by using stochastic dynamic operators (SDOs). These new methods have potential to significantly improve predictions of dynamics from recordings of brain function, which in turn would have significant technological and medical impacts in areas including disease process diagnosis, disease control using stimulation, robot prostheses and brain machine interface designs, neural prostheses, and neurally-driven augmentation or replacement. The project brings together a collaboration between an applied mathematician/neurologist and a comparative neurophysiologist, and will provide interdisciplinary graduate and postdoctoral training at the cutting edge of neuroscience, stochastic methods and control.
Increasing sophistication of brain recording technology is not fully matched by an equally sophisticated mathematical approach that permits modeling and direct prediction of the relation between behavior and the activity of neural populations. For motor systems, the primary goal is control of dynamics in the environment. The methods under investigation avoid the usual neural separation into sensory and motor effects. They treat neural activity as representing probabilistic alterations of unfolding dynamics. More specifically, the SDO framework considers neural activity as causing a modification of the overall system dynamics, so that the resulting dynamics (including movement, compliance, and oscillatory behavior) achieve a desired result. This allows principled engineering solutions and use of 'big' neural activity to predict dynamics. The proof of concept proposal will test model prediction responses during trajectory formation and perturbation in reflex behavior, prediction of real-time effect of single spikes, and combined effect of multiple neurons/populations. On proof of concept project completion: (1) The SDO framework will be compared with classical techniques using novel data sets; (2) Basic feasibility of real-time robot control from spinal neural activity in a model system will be assessed. Together, these data will all add significantly to neural analysis, neurotechnology and understanding of the novel methods in relation to others.
