Funded Grants

How does the brain learn movement? Bridging the gap between behavioral processes and functional imaging signals

From the fluid grace a of ballet dancer to the lightning finger-speed of a concert pianist, the ability to produce skillful movements is one of man's most astonishing capabilities. Motor control, however, is a hard problem -- complex computations are necessary to control even the most mundane of movements. One can appreciate this fact when noting that even advanced robotic devices only poorly approximate human performance, or when watching a stroke victim try to relearn motor skills once taken for granted.

What happens in the human brain when we learn new motor skills? Because recovery from stroke and healthy motor learning are likely driven by overlapping mechanisms, this question is not only important to basic neuroscience, but also has profound clinical implications. Functional magnetic resonance imaging (fMRI) is one of the most powerful techniques for studying activity in the human brain and ought to play a key role in answering this question. However, despite many efforts over the past two decades fMRI studies have contributed disappointingly little to our understanding of motor learning. Hundreds of published studies have resulted in a bewildering and often inconsistent picture of activation changes, without providing a satisfying answer to what "more" or "less" activity means. I believe that the main impediment in the field is that we lack explicit theories that that tell us what fMRI signal we should observe when a region undergoes learning.

The focus of my research program is therefore to develop this missing link between motor learning and signals observable in fMRI. Such a theory has two requirements: First, we need a computational description of the neural learning processes. It has become clear that motor learning is not monolithic, but consists of a number of interacting mechanisms. For example, the motor system learns from various forms of reinforcement, from the small errors that occur during motor execution, and from simply performing the same movement repeatedly. These mechanisms likely rely on different plasticity processes and therefore have their own neural signatures. Second, we need a theoretical framework that predicts the form of the neural representations that result from learning processes. What distinguishes trained movements from untrained movements? Should we expect more activation? A larger area of activation? I have developed a modeling approach that suggests a novel way of detecting motor skill representations in distributed patterns of activity.

My research program aims to develop a quantitative theory that bridges the gap between task-level descriptions and observable fMRI signals. I will test these ideas in healthy human brain using stimulation techniques to influence learning processes, and in stroke recovery. By using fMRI to observe in vivo changes in the human brain, these developments will accelerate the development of new diagnostic techniques, and ultimately, guide our search for more effective rehabilitation techniques.