Thesis
Categorical models in control theory & reinforcement learning
- Creator
- Rights statement
- Awarding institution
- University of Strathclyde
- Date of award
- 2026
- Thesis identifier
- T17637
- Person Identifier (Local)
- 202163348
- Qualification Level
- Qualification Name
- Department, School or Faculty
- Abstract
- In the scientific inquiry, the integration of disparate fields from innocent connections often yields new perspectives. This thesis explores the bidirectional nature of some methods in Control Theory and Reinforcement Learning. The languages in which these problems are stated are varied as they seek to address different questions, while engaging with common structures. Our aim is to illuminate some bridges between these applied fields, and we believe that category theory is the right language for this. Reinforcement learning (RL) refers to a class of methods in machine learning for optimising a long-run reward during interaction with an unknown environment. It is considered one of the major pillars of machine learning, along with deep learning (neural networks and differentiable programming), unsupervised learning (statistical clustering methods, including topological data analysis [72]) and variational learning (Bayesian inference and related probabilistic methods). It can be seen as an extension of dynamic programming methods in optimal control theory [24], which drops the assumption that a model of the environment is known. RL, combined with deep learning methods to produce deep RL, notably achieved state of the art success in practical game playing, with AlphaGo defeating the human Go champion in 2016 and AlphaStar achieving Grandmaster status in the real time strategy game StarCraft II.
- Advisor / supervisor
- Hedges, Jules
- Ghani, Neil
- Resource Type
- DOI
Relations
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
PDF of thesis T17637 | 2026-04-15 | Public | Download |