Vincent S. Huang

Learning to control a new tool (i.e., a novel environment) produces an internal model, i.e., a motor memory that allows the brain to implicitly predict the behavior of the tool. Data from a wide array of experiments suggest that formation of motor memory is not a single process, but one that is due to multiple adaptive processes with different time constants. Here we asked whether these time constants are invariant or are they influenced by the statistics of the learning event. To measure the time constants, we controlled the statistics of the learning event in a reaching task and then assayed the decay rates of motor output in a set of trials in which errors were effectively removed. We found that prior experience with a rapid change in the environment increased the decay rate of memories acquired later in response to a gradual change in the same environment. Prior experience in an environment that changed gradually reduced the decay rates of memories acquired later in response to a rapid change in that same environment. Indeed we found that by manipulating the prior statistics of the learning experience, we could readily alter the decay rates of a given motor memory. This suggests that time scales of processes that support motor memory are not constant. Rather decay of motor memory is the brain's implicit estimate of how likely it is that the environment will change with time. During motor learning, prior statistics that suggest changes are likely to be permanent result in slowly decaying memories, whereas prior statistics that suggest changes are transient result in rapidly decaying memories.

Conventional neurorehabilitation appears to have little impact on impairment over and above that of spontaneous biological recovery. Robotic neurorehabilitation has the potential for a greater impact on impairment due to easy deployment, its applicability across of a wide range of motor impairment, its high measurement reliability, and the capacity to deliver high dosage and high intensity training protocols.

When we learn a new skill (e.g., golf) without a coach, we are “active learners”: we have to choose the specific components of the task on which to train (e.g., iron, driver, putter, etc.). What guides our selection of the training sequence? How do choices that people make compare with choices made by machine learning algorithms that attempt to optimize performance? We asked subjects to learn the novel dynamics of a robotic tool while moving it in four directions. They were instructed to choose their practice directions to maximize their performance in subsequent tests. We found that their choices were strongly influenced by motor errors: subjects tended to immediately repeat an action if that action had produced a large error. This strategy was correlated with better performance on test trials. However, even when participants performed perfectly on a movement, they did not avoid repeating that movement. The probability of repeating an action did not drop below chance even when no errors were observed. This behavior led to suboptimal performance. It also violated a strong prediction of current machine learning algorithms, which solve the active learning problem by choosing a training sequence that will maximally reduce the learner's uncertainty about the task. While we show that these algorithms do not provide an adequate description of human behavior, our results suggest ways to improve human motor learning by helping people choose an optimal training sequence.