Neural population dynamics in motor control and learning
File(s)
Author(s)
Chang, Chen Hsuan
Type
Thesis or dissertation
Abstract
Animals have a remarkable ability to learn and control a wide range of movements. This ability is crucial not only for the survival of an individual over its lifetime, but also for the survival of its species over evolutionary timescales. How the brain achieves both the stability required for precise control and the flexibility required for learning new skills remains an open question. Recent work in motor neuroscience has found that neural population activity is largely confined to a low-dimensional subspace called the neural manifold. Examining the time-varying activity in this space—the population latent dynamics—has provided many insights into the neural underpinnings of movement. In this thesis, we examine the role of latent dynamics in motor control and learning in three parts.
First, we explore the stability of latent dynamics in motor control by asking how different individuals of the same species are able to generate similar behaviors. We hypothesize that evolutionarily conserved latent dynamics allow animals to reliably produce the same adaptive behaviors. Through a combination of experimental recordings and computational modeling, we demonstrate that latent dynamics are consistently stable and preserved across multiple motor domains. Second, we investigate the flexibility of latent dynamics during motor learning. Using recurrent neural networks to model motor cortical dynamics, we examine how population activity acquired through de novo skill learning influences the flexibility of subsequent motor adaptation. Third, we explore how animals can flexibly learn new motor tasks without disrupting stable motor control. By modeling a sequential learning experiment, we investigate neural signatures of motor memories and their relationship to continual learning.
Overall, our work establishes the importance of latent dynamics in motor control and motor learning, offering new perspectives on how the brain balances stability and flexibility for behavior, across an animal's lifetime and over the course of a species’ evolution.
First, we explore the stability of latent dynamics in motor control by asking how different individuals of the same species are able to generate similar behaviors. We hypothesize that evolutionarily conserved latent dynamics allow animals to reliably produce the same adaptive behaviors. Through a combination of experimental recordings and computational modeling, we demonstrate that latent dynamics are consistently stable and preserved across multiple motor domains. Second, we investigate the flexibility of latent dynamics during motor learning. Using recurrent neural networks to model motor cortical dynamics, we examine how population activity acquired through de novo skill learning influences the flexibility of subsequent motor adaptation. Third, we explore how animals can flexibly learn new motor tasks without disrupting stable motor control. By modeling a sequential learning experiment, we investigate neural signatures of motor memories and their relationship to continual learning.
Overall, our work establishes the importance of latent dynamics in motor control and motor learning, offering new perspectives on how the brain balances stability and flexibility for behavior, across an animal's lifetime and over the course of a species’ evolution.
Version
Open Access
Date Issued
2025-04-07
Date Awarded
2026-03-01
Copyright Statement
Attribution-Non Commercial-No Derivatives 4.0 International Licence (CC BY-NC-ND)
Advisor
Clopath, Claudia
Gallego, Juan Álvaro
Sponsor
Wellcome Trust (London, England)
Grant Number
108908/Z/15/Z)
Publisher Department
Department of Bioengineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)