Probabilistic approaches to statistical and structure learning in human cognition

Silvestrin, Francesco (2022) Probabilistic approaches to statistical and structure learning in human cognition. Doctoral thesis, University of East Anglia.

Preview

PDF Download (9MB) | Preview

Abstract

This thesis describes a series of empirical and simulation experiments, all in the broad area of probabilistic inference and learning. The first three experiments, described in Chapters 4 and 5, focus on a specific theoretical framework, predictive coding. We identified some critical issues (discussed in more detail in Chapter 1) and tackled them with a combination of techniques including pupillometry, electroencephalography (EEG) and computational modelling. In particular, we present an augmented version of classical predictive coding models incorporating dynamic precision estimation (Chapter 4) and show how human participants can successfully learn multimodal distributions, violating classical predictive coding (Chapter 5). In Chapter 6 we took a more theoretically agnostic approach (although we firmly remained within the Bayesian brain framework) to study structure learning. If in Chapter 5 we verified that humans could learn multimodal distributions, in Chapter 6 we asked how they do it without having any knowledge about the structure of the probabilistic model generating their observations. We also introduce a working memory component, with our simulations showing how revisiting past stimuli can benefit structure learning.

Overall, we contribute to the Bayesian brain framework both with empirical findings coming from both simulations and lab experiments. We augment current computational models increasing their flexibility, and thus their scope to be used in more diverse experimental contexts. Finally, we make a contribution to the field of computational rationality, discussing the trade off between working memory load and learning performance.

Item Type:	Thesis (Doctoral)
Faculty \ School:	Faculty of Social Sciences > School of Psychology
Depositing User:	Nicola Veasy
Date Deposited:	26 Oct 2023 08:18
Last Modified:	26 Oct 2023 08:18
URI:	https://ueaeprints.uea.ac.uk/id/eprint/93476
DOI:

Downloads

Actions (login required)

View Item