Structure Learning and learning structure with Active Inference

Advanced search
Browse by:

Department | Year

UCL Theses | Latest

Deposit your research

Structure Learning and learning structure with Active Inference

Neacsu, Victorita Oana; (2024) Structure Learning and learning structure with Active Inference. Doctoral thesis (Ph.D), UCL (University College London). Green open access

Preview

Text
Victorita Oana Neacsu - Structure Learning and learning structure with Active Inference.pdf - Other
Download (5MB) | Preview

Abstract

This thesis is based on the Active Inference Framework (AIF), a theoretical model of information processing that views agents as inference machines. While a vast amount of research within the AIF focuses on inference and associative learning, Structure Learning (SL) is a newer and less established aspect of the AIF landscape. This thesis aims to clarify Structure Learning through three main lines of inquiry: defining Structure Learning, illustrating its implementations, and offering evidence for this construct’s alignment with human behaviour. The thesis starts with a synthesis of structure learning in the general literature from research in humans, ethology, and in silico (Chapter 1). Chapter 2 will introduce three main levels of information processing in the AIF (Active Inference, Parametric Learning, and Structure Learning), and shows how various features of SL – from the general literature – relate to SL as implemented in the AIF. Chapter 3 will showcase computational simulations of a geocaching task using a deep AIF model. We show that synthetic agents learn the environmental structure through Active Inference and Parametric Learning, resulting in two types of foraging: goaldirected navigation and epistemically driven exploration. In Chapter 4, a deep hierarchical AIF model is employed to elucidate how SL influences concept learning. When endowed with SL, synthetic agents show improved performance during spatial foraging: they accumulate more rewards and show higher information gain. Chapter 5 illustrates the learning of a more abstract type of structure: learning about regularities in the environment in the form of abstract rules that underlie observed outcomes. The work in this chapter is the first to date to show evidence for Structure Learning (as implemented in the AIF) in a cognitive task in humans. In Chapter 6, I will briefly recapitulate the findings, discuss their implications, and suggest possible future directions.

Type:	Thesis (Doctoral)
Qualification:	Ph.D
Title:	Structure Learning and learning structure with Active Inference
Open access status:	An open access version is available from UCL Discovery
Language:	English
Additional information:	Copyright © The Author 2024. Original content in this thesis is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) Licence (https://creativecommons.org/licenses/by-nc/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. Access may initially be restricted at the author’s request.
UCL classification:	UCL UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Brain Sciences UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Brain Sciences > UCL Queen Square Institute of Neurology
URI:	https://discovery.ucl.ac.uk/id/eprint/10192997