Motor learning in the saccadic system: The importance of prediction in maintaining movement accuracy.

摘要

One primary way in which the brain maintains movement accuracy is via error-based motor learning. This is thought to involve a prediction-based process, which drives adaptation to resolve unexpected errors. This dissertation investigates the role of prediction in motor learning. In particular, I studied the dependence of adaptation on a prediction-based error signal, explored the connection between motor prediction tasks and adaptation, investigated the role of the cerebellum in this prediction-based learning process, and developed a computational model to gain insight into this prediction-based learning process. Studies involved recording saccadic eye movements while subjects were presented with tasks designed to specifically address the role of prediction in motor learning. Both normal control subjects as well as patients with spinocerebellar ataxia type 6 were studied. Results indicate that adaptation depends heavily upon a prediction-based error signal to drive learning. Motor prediction tasks exhibit similar characteristics to adaptation, including direction-specificity and multiple time scales of learning. Adaptation can also be directly driven by a purely prediction-based task, suggesting shared neural machinery underlying both processes. With cerebellar impairment, there is a fundamental change in how patients respond to a prediction-based adaptation task, indicating that the cerebellum is critical for prediction-driven error-based motor learning. Modeling results confirm that prediction and adaptation are related, and the way in which current and past information is used to drive future movement corrections influences the observed fluctuations in behavior that are often assumed to be random noise. Finally, a follow-up experiment based on model implications confirms that prediction and adaptation are closely related tasks in that they seem to rely on a shared -- perhaps fractal -- learning process. This finding suggests a novel interpretation for observed fractal fluctuations in behavior: they reflect a learning system response that is a balance of rapid responses to errors and stable long-term performance. In conclusion, prediction has a prominent role in maintaining movement accuracy and is critical for error-based motor learning.