My current work at Brown with Michael Black focuses on developing mathematical methods for decoding neural (cortical) code for movement and using it for direct cortical control of motor activity in artificial systems. The primary application for this is in neuro-motor prosthetics for patients whose motor cortex is intact but who have lost control control of motor function due to injury or decease. We would like to bypass the damaged neural pathways by means of computation: decode the "commands" issued by the brain and translate them into commands to, say, a robotic manipulator. Of course, such understanding of the neural code would also have great scientific implications. We are working in collaboration with Donoghue Lab and Cyberkinetics, Inc.

My primary interests in this area are related to statistical learning methods that address the following problems.

• Motor primitives in movement and in neural code for dexterous hand maipulation. Is there an inherent low-dimensional structure in the high-dimensional space of possibe hand movements? How is this structure related to the neural signal recorded in motor cortex? This requires developing new efficient methods of unsupervised and semi-supervised learning, if we want to reason about temporal structure, and not just the space of feasible static postures.

  With colleagues from Donoghue Lab in Brown Neuroscience we are currently conducting experiments in which neural signal and natural complex behavior are recorded simultaneously. We also have been collecting and analyzing data from human subjects performing complex manipulations.

• How can we measure the relevance of neural signal to a particular behavior (or, alternatively, to a stimulus - a question that arises in computational analysis of sensory, rather than motor, mechanisms)? This is not the same as decoding--asking how the behavior in question is encoded in the neural signal.

  What makes this particularly challenging is the high dimensionality of the neural signal when recorded with a multi-electrode array. Straightforward non-parametric methods for assessing mutual information break down in this domain.

• The natural motor-related mechanisms in the brain have evolved to control certain biomechanical systems, such as the primate arm and hand. In the context of neuro-motor prosthetics the same mechanisms have to control an artificial system, such as a robotic manipulator, a wheelchair or, in the simplest case, a computer cursor. I am interested in developing models that allow efficient decoding and robust control for this application. One interesting direction is to incorporate physical constraints inherent in the real world directly into the decoding process. Some preliminary results of work done with Phil Kim and Michael Black were presented at NIPS 2006 (also see our SfN 2006 poster).