Understanding human motor control and learning - a mathematical perspective

人类是如何学会控制身体来完成运动任务的？这个看似微不足道，但却至关重要的问题仍然悬而未决。已经提出了许多理论，例如，自下而上的方法（使用第一原理的生理建模）和自上而下的框架（使用统计数据驱动的建模方法）。然而，上述两种方法在很大程度上是分开的，它们之间的相互作用和集成——这本来可以解决这两种方法的一些关键限制——是缺失的。该项目旨在利用机器学习（特别是可解释的深度强化学习）和智能控制系统（特别是迭代学习控制）的最新发展来弥合这一差距，以创造一种新的理论来解释人类运动控制和学习。支撑这一理论的是优化方法的最新进展，特别是近年来对大数据研究和新的人工智能算法的需求激增，推动了非常高维的问题。这个跨学科的项目将提高神经科学家对人类大脑如何学习和控制身体运动任务的理解。这种理解将进一步反馈到机器学习和智能控制的研究中，使我们能够开发更有效的算法。这种更好的理解也将有助于开发更好的康复技术，帮助患者以更自然、更有效的方式重新学习他们的运动。

How does human learn to control the body to perform a motion task? This seemingly trivial, but fundamentally important question, remains open. A number of theories have been proposed, e.g., the bottom-up approach (using physiological modelling from first principles) and top-down framework (using statistical data driven modelling methods). However, the aforementioned two approaches are largely separated, the interaction and integration of which - that could have addressed a number of key limitations in both methods - is missing. This project aims to bridge this gap exploiting the recent developments in machine learning (particularly in interpretable deep reinforcement learning) and intelligent control systems (especially iterative learning control) to create a novel theory to explain human motor control and learning. Underpinning this theory is the recent progress in optimisation methods, particular for very high dimensional problems that have been driven by recent years' surging demand of big data research and new AI algorithms.This interdisciplinary project will improve the neuroscientist's understanding of how human brain learns and controls the body for motion tasks. This understanding will further feedback into the research of machine learning and intelligent control, enabling us to develop more efficient algorithms. This improved understanding will also help to develop better rehabilitation technologies helping patients re-learn their motion movements in a more natural and effective way.