Empirical studies on the computational and cognitive mechanisms of human learning and movement

Learning;Movement;Cognition;Human behavior;Biomedical Engineering

Huberdeau, David MBastian, Amy ;

Johns Hopkins University

关键词: Learning;    Movement;    Cognition;    Human behavior;    Biomedical Engineering;   

Others  :  https://jscholarship.library.jhu.edu/bitstream/handle/1774.2/60190/HUBERDEAU-DISSERTATION-2017.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: JOHNS HOPKINS DSpace Repository

【 摘 要 】

The topic of human movement, and the question of how humans learn new behaviors, has puzzled philosophers and scientists since classical times. A commonly held assumption is that there are two qualitatively distinct learning systems, one responsible for remembering knowledge of facts and events, and the other responsible for forming associations and learning new skills, including motor learning. The evidence in support of this dissociation has been independently reproduced through many different experiments and methods of analysis. One line of evidence that has recently been investigated is the dual-component nature of adaptation learning. When humans and animals are challenged with a change in their environment or the physiology of their bodies, such as what might happen through growth and development or because of injury, the nervous system adjusts its control mechanisms to maintain accurate movements. Learning of this form is known as adaptation, and had originally been theorized to be achieved through an implicit learning mechanism. Furthermore, it was often thought that this same learning mechanism was responsible for more general forms of learning, such as learning the use of new tools. This model has recently come under scrutiny as evidence has emerged demonstrating a role for memory of facts in adaptation. If there are at least two mechanisms responsible for adaptation learning, which one of them, if either, is actually responsible for more general skill learning? If one, but not the other, of these mechanisms is responsible for skill learning, what is adaptation really a model of? And how might the conclusions of other studies that used adaptation as a general model for learning need to be reconsidered? For instance, the results from neurophysiological studies of adaptation may find neural correlates that are uniquely related to adaptation but not to other types of motor learning. Having a better behavioral- and computational-level understanding of the mechanisms involved in adaptation learning is necessary to address these and potentially many other questions. Given the challenges present in the study of adaptation, there is a need for other models of learning and movement that give different perspectives and emphasize other aspects of learning that might be missing from adaptation. For instance, adaptation involves correction of movements around an existing ability, such as reaching. How is reaching itself learned? Acquiring or building new behavioral abilities might involve qualitatively different mechanisms compared to adaptation. Furthermore, new methods for analyzing the kinematics of movements are necessary, as adaptation paradigms typically limit their analysis to the choice of reaching direction only. In this dissertation, I will present several original, empirical studies on the role of cognition and explicit knowledge in motor learning. I will investigate the computational mechanisms that underlie learning new behaviors. I will introduce a new model for human motor skills and skill learning, and show how this model fills gaps that exist in the repertoire of models, methods, and concepts currently popular in the science of learning. I will show evidence that adaptation learning is made up of at least two qualitatively distinct learning components. One component appears to be deliberate, driven by explicit knowledge, and is computationally expensive. The other is implicit, driven by sensory-prediction errors, and is automatic and readily expressed. I will demonstrate that the deliberate component becomes automatic following practice, and will argue that this process is a plausible mechanism for how more general motor skills are learned. Implicit recalibration does not change with practice and therefore appears unlikely to be responsible for skill learning. I will show that learning a new continuous-movement behavior, like skiing or riding a bike, is done through the creation of a flexible feedback control policy. I will discuss the inconsistency of sequence learning and chunking hypotheses, and contrast them with the control policy theory. The studies, results, and conclusions presented here demonstrate that motor learning intrinsically involves cognition and explicit representations of knowledge. The classical concept of motor learning being a subset of implicit memory is inconsistent with the present findings and other recent work. Instead, a view of motor learning as being a phenomenon emergent from the interaction of multiple forms of memory and algorithms of learning is emerging.