Computations in Human Motor Learning

人类运动学习中的计算

• 批准号: 10560638
• 负责人: Daniel Wolpert
• 金额: $ 49.14万
• 依托单位: COLUMBIA UNIVERSITY HEALTH SCIENCES
• 依托单位国家: 美国
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-02-15 至 2026-01-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10560638
• 关键词: 3-Dimensional; Address; Affect; Appearance; Belief; Biological Assay; Brain; Cues; Custom; Data; Degenerative Disorder; Development; Disease; Environment; Experimental Designs; Family; Feedback; Geometry; Goals; Golf; Height; Human; Impairment; Individual; Knowledge; Laboratories; Lead; Learning; Length; Longevity; Maps; Memory; Modeling; Modification; Motion; Motor; Movement; Muscle; Nervous System; Outcome; Participant; Persons; Play; Probability; Process; Property; Racquet Sports; Recording of previous events; Recovery; Rehabilitation therapy; Research; Robotics; Role; Savings; Sensory; Squash; Stream; Stroke; Tennis; Testing; Texture; Time; Update; Visual; Weight; Work; arm; arm movement; behavioral study; computer framework; developmental disease; experience; improved; innovation; motor behavior; motor control; motor learning; nervous system disorder; neural; novel; sensory input; statistics; tool; virtual reality

Project Summary The long-term goal of our laboratory is to understand the computations underlying human motor learning and thereby provide a framework to examine the neural underpinnings of learning, the deficits seen in neurological disorders and how learning mechanisms can be leveraged in rehabilitation. Motor learning is the fundamental process that involves changes in motor behavior arising from interaction with the environment. Humans spend a lifetime learning, storing and refining a multitude of motor memories appropriate for different contexts. Current studies of motor learning have focused almost exclusively on adaptation of individual memories in isolation. Con- sequently, the principles underlying how the brain coordinates its repertoire of memories are largely unknown. Our key hypothesis is that the process of contextual inference, estimating the probability with which each exist- ing motor memory is appropriate for the current situation, controls the creation of new memories and the degree to which different memories are expressed and updated. Our objective is to understand what leads to the cre- ation of new memories compared to the modification of existing motor memories, and how existing memories are recalled and updated. We have developed the COIN (COntextual INference) model to formalize the role of contextual inference in motor learning. The COIN model performs contextual inference in a more principled and comprehensive way than any previous model and can explain key findings traditionally attributed to adaptation as arising instead from contextual inference, such as spontaneous recovery, savings, anterograde interference and changes in learning rates. In contrast to current models, a critical feature of the COIN model is that it can determine, in a principled manner, whether a new memory should be created or existing memories adapted. To both test and develop the model, we will use behavioral studies in humans using novel robotic interfaces and virtual reality which allow us to control a participant’s sensorimotor experience during motor learning tasks. In Aim 1 we will determine the conditions under which new motor memories are created. In Aim 2 we will determine the rules by which existing motor memories are updated. While Aims 1 and 2 focus on reaching movements in the plane which make a large body of previous research comparable, Aim 3 moves towards more naturalistic tasks of manipulating objects in three-dimensions. In Aim 3 we will determine how motor memories are organized into fam- ilies to allow efficient learning and generalization for contexts that share similar properties. Voluntary movement is fundamental to human existence, yet many diseases such as stroke, degenerative disease, and developmental disorders, impair human movement over the life span. By establishing a new framework of motor learning, this project will contribute to our ultimate goal of developing assays to understand deficits in neurological disorders and develop paradigms that can control the contextual inference process so as to improve rehabilitation.

项目概要 我们实验室的长期目标是了解人类运动学习和 提供一个框架来检查学习的神经基础，从而在神经学中看到的缺陷 障碍以及如何在康复中利用学习机制是基础。 涉及人类与环境相互作用引起的运动行为变化的过程。 一生学习、存储和提炼适合不同环境的大量运动记忆。 运动学习的研究几乎完全集中在孤立的个体记忆的适应上。 因此，大脑如何协调其记忆库的基本原理在很大程度上尚不清楚。 我们的关键假设是上下文推理的过程，估计每个存在的概率 - 运动记忆适合当前情况，控制新记忆的创建和程度 我们的目标是了解是什么导致了不同的记忆的表达和更新。 新记忆与现有运动记忆的修改相比，以及现有记忆如何 我们开发了 COIN（情境推理）模型来形式化角色的作用。 运动学习中的上下文推理 COIN 模型以更有原则性和更有效的方式执行上下文推理。 比任何以前的模型都更全面，并且可以解释传统上归因于适应的关键发现 来自上下文推理，例如自发恢复、储蓄、顺行干扰 与当前模型相比，COIN 模型的一个关键特征是它可以 以原则性的方式确定是否应该创建新的记忆或改编现有的记忆。 测试和开发模型，我们将使用新颖的机器人接口对人类进行行为研究， 虚拟现实使我们能够在运动学习任务中控制参与者的感觉运动体验。 1 我们将确定创建新运动记忆的条件 在目标 2 中，我们将确定创建新运动记忆的条件。 更新现有运动记忆的规则，而目标 1 和 2 侧重于实现运动。 与之前的大量研究相比，Aim 3 朝着更自然的任务迈进 在目标 3 中，我们将确定运动记忆如何组织成家族。 有助于对具有相似属性的环境进行有效的学习和概括。 是人类生存的基础，但许多疾病，如中风、退行性疾病和发育性疾病 通过建立新的运动学习框架，这会损害人类一生的运动。 该项目将有助于我们的最终目标，即开发分析方法来了解神经系统疾病的缺陷 并开发可以控制上下文推理过程的范式，以改善康复。