Landscape Models of Motor Learning

Gottfried Mayer-Kress, Department of Kinesiology, Penn State University

Whereas we have many senses to receive information, we always have to initiate coordinated muscle activity if we want to interact with our environment. In our daily life simple activities like standing, or walking require the synergistic cooperation of a large number of neuronal and muscle cells. Most, if not all of these coordinated activities need to be learned and our performance improves gradually (e.g. when we try to learn to play golf) or in critical transitions (e.g. when we learn to ride a bicycle).
It has been common experience that our performance in any given task improves with practice. The study of learning curves has a history of more than a hundred years and one of the main outcomes was a universal power-law of practice. We re-analyzed historical learning data and could show that many if not all learning curves are only poorly approximated by a power law but that they reveal structures that we claim can be better described by a finite number of discrete time-scales with the power-law being contained as a special case. By reconstructing a performance landscape from the learning curve we could show that different time-scales are affected differently by rest periods without practice. In some dimensions rest/sleep will lead to a temporary decrease in performance whereas in other dimensions sleep can actually improve performance. We claim that this theoretical interpretation might resolve long-standing disputes surrounding the role of sleep in learning. \

Slides Audio