Humans possess a remarkable ability to learn motor tasks very quickly. Recent findings are beginning to discover that this ability is supported by multiple learning systems, each with their unique computational features. The goal of this research proposal is to characterize these learning processes at a computational and neural systems level. The signatures of at least three separate learning mechanisms have been identified. Reinforcement learning can relate task success with a particular action to better guide selection among competing action alternatives. A forward model can learn to make predictions about the sensory consequences of a selected action to improve motor execution. While these two processes are capable of refining performance, one based on reinforcement and the other from movement error, they entail a relatively slow, gradual process requiring extensive practice. A striking feature of human competence is the ability to employ explicit strategies that can facilitate learning at a much faster time scale. We hypothesize that these three processes are dependent on distinct neural circuits that can operate concurrently during a motor learning task. Each may operate with a significant degree of independence, yet they can, in certain circumstances, interact to converge of a common learning solution. We will exploit a simple learning task involving a visuomotor rotation to parametrically manipulate the relative contribution of these processes to motor learning. We propose that changes in task conceptualization and feedback are critical in determining the relative engagement of these processes. Empirical tests and computational modeling will provide a rigorous analysis of these hypotheses and to develop a process-based account of motor learning. The neuroanatomical substrates of these processes will be explored by testing neurological populations with degenerative disorders of the cerebellum or basal ganglia, and patients with cortical lesions affecting prefrontal regions. We assume that reinforcement learning, forward model adaptation, and strategy-based control are likely operative in nearly all motor learning tasks, but their individual contribution to learning has largely been overlooked. If learning is the weighted contribution of these processes, then differences in task information as well as individual differences in the exploitation of this information, will influence the relative contribution to learning, even when the basic task remains unchanged. Ultimately, understanding how multiple systems contribute to learning should lead to the development of optimal training protocols either designed to target impaired systems or bias performance to rely on systems that are relatively intact.
Stroke and neurodegenerative diseases can lead to a profound loss of motor function, significantly impairing daily activities. The N.I.H. has identifie the need to develop novel training methods to restore or replace lost function in neurologically disabled individuals in its report Neuroscience in the New Millennium. This project aims at examining how different processes, which are reliant on distinct neural circuits, can contribute independently but in an integrated manner to the learning of motor skills. Characterization of these processes, as well as defining tasks that differentially exploit these processes provides an important step towards designing neurorehabilitation programs that are tailored to favor learning by systems that remain functional.
|Taylor, Jordan A; Krakauer, John W; Ivry, Richard B (2014) Explicit and implicit contributions to learning in a sensorimotor adaptation task. J Neurosci 34:3023-32|
|Taylor, Jordan A; Ivry, Richard B (2014) Cerebellar and prefrontal cortex contributions to adaptation, strategies, and reinforcement learning. Prog Brain Res 210:217-53|