Neural correlates of movement planning under uncertainty
File(s)
Author(s)
Bagi, Bence
Type
Thesis
Abstract
Movement is a key aspect of daily life, serving as the primary means by which animals, including humans, interact with their environment.
Here, I investigate how uncertainty shapes neural activity during movement planning and preparation, focusing on the primary motor cortex (M1) and dorsal premotor cortex (PMd) of nonhuman primates. By examining covariability in neuronal populations and analyzing their activity through low-dimensional neural subspaces, I demonstrate that when animals must plan and execute reaches under uncertain visual conditions, PMd and M1 exhibit distinct yet complementary roles. Specifically, while information about both uncertainty and reach direction is present in both regions, PMd prominently encodes uncertainty, aligning it with the dominant dimensions of neural population activity, whereas M1 primarily represents reach direction. Furthermore, I show that uncertainty and reach direction occupy orthogonal subspaces within the population activity space. These findings suggest a functional gradient across these regions of the motor cortex when generating movement under uncertainty.
Building on these results, I employ multi-module neural network models trained under both supervised and reinforcement learning paradigms. These models provide insights into how certain experimentally observed features might arise from network dynamics, architectural constraints, and distinct computational roles assigned to different modules. Additionally, fitting recurrent neural networks directly to recorded firing rates illustrates that the patterns observed in experimental recordings can be generated from the given task inputs via learned dynamics.
Extending the analysis to additional datasets – one from monkeys and another from mice – performing sensory-based decision-making tasks reveals a similar functional gradient, supporting the generality of the observed organizational principles.
Together, these findings advance our understanding of how neural circuits integrate uncertainty into movement planning and execution.
Here, I investigate how uncertainty shapes neural activity during movement planning and preparation, focusing on the primary motor cortex (M1) and dorsal premotor cortex (PMd) of nonhuman primates. By examining covariability in neuronal populations and analyzing their activity through low-dimensional neural subspaces, I demonstrate that when animals must plan and execute reaches under uncertain visual conditions, PMd and M1 exhibit distinct yet complementary roles. Specifically, while information about both uncertainty and reach direction is present in both regions, PMd prominently encodes uncertainty, aligning it with the dominant dimensions of neural population activity, whereas M1 primarily represents reach direction. Furthermore, I show that uncertainty and reach direction occupy orthogonal subspaces within the population activity space. These findings suggest a functional gradient across these regions of the motor cortex when generating movement under uncertainty.
Building on these results, I employ multi-module neural network models trained under both supervised and reinforcement learning paradigms. These models provide insights into how certain experimentally observed features might arise from network dynamics, architectural constraints, and distinct computational roles assigned to different modules. Additionally, fitting recurrent neural networks directly to recorded firing rates illustrates that the patterns observed in experimental recordings can be generated from the given task inputs via learned dynamics.
Extending the analysis to additional datasets – one from monkeys and another from mice – performing sensory-based decision-making tasks reveals a similar functional gradient, supporting the generality of the observed organizational principles.
Together, these findings advance our understanding of how neural circuits integrate uncertainty into movement planning and execution.
Version
Open Access
Date Issued
2024-12-20
Date Awarded
2025-10-01
Copyright Statement
Attribution-NonCommercial 4.0 International Licence (CC BY-NC)
License URL
Advisor
Gallego, Juan Alvaro
Publisher Department
Department of Bioengineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)