From thought to action

Abstract

Systems engineering is rapidly assuming a prominent role in neuroscience that could unify scientific theories, experimental evidence, and medical development. In this three-part work, I study the neural representation of targets before reaching movements and the generation of prosthetic control signals through stochastic modeling and estimation. In the first part, I show that temporal and history dependence contributes to the representation of targets in the ensemble spiking activity of neurons in primate dorsal premotor cortex (PMd). Point process modeling of target representation suggests that local and possibly also distant neural interactions influence the spiking patterns observed in PMd. In the second part, I draw on results from surveillance theory to reconstruct reaching movements from neural activity related to the desired target and the path to that target. This approach combines movement planning and execution to surpass estimation with either target or path related neural activity alone. In the third part, I describe the principled design of brain-driven neural prosthetic devices as a filtering problem on interacting discrete and continuous random processes. This framework subsumes four canonical Bayesian approaches and supports emerging applications to neural prosthetic devices.

(cont.) Results of a simulated reaching task predict that the method outperforms previous approaches in the control of arm position and velocity based on trajectory and endpoint mean squared error. These results form the starting point for a systems engineering approach to the design and interpretation of neuroscience experiments that can guide the development of technology for human-computer interaction and medical treatment.