| Our seemingly effortless ability to make coordinated movements belies the sophisticated computational machinery at work in our nervous system. Much has been learned about motor cortical processing with classic systems neuroscience approaches. In recent years, the field has been dramatically expanding the complexity of its data, recording technologies, and experiments. This shift seeks to deliver a much deeper understanding of cortical processing and a much improved ability to control neural prosthetic devices (also called brain-machine interfaces). Realizing this payoff, however, requires analytical and computational methods that can exploit this changing paradigm. This dissertation describes algorithmic developments for understanding cortical processing and for prosthetic systems. The first part focuses on our signal processing efforts to extract useful signals underlying noisy neural activity in the motor system of the primate brain, both for single neurons and for populations of simultaneously recorded neurons. Specifically, I discuss analyzing single neurons using machine learning techniques (Gaussian Processes) in a point process framework, and I describe an approach to mitigate the significant optimization challenges of this method. I then discuss extending this idea across many simultaneously recorded neurons (with a factor analysis-like algorithm) to extract population-level signatures of neural activity. This part also points to broader implications of this work for engineering and statistics. The second part focuses on algorithmic challenges in neural prosthetic systems, first describing performance improvements available via algorithmic optimization of a prosthetic interface. I then discuss other algorithmic areas of prosthetic design, reviewing a number of areas for improvement and pointing to future work to address these current problems. As a whole, this dissertation offers analytical methods that push forward advanced research into the brain's motor system and our ability to meaningfully interface with it. |
