Anxiety and computations of uncertainty during reward-based learning in volatile environments

Hein, Thomas. 2021. Anxiety and computations of uncertainty during reward-based learning in volatile environments. Doctoral thesis, Goldsmiths, University of London [Thesis]

[img]
Preview
Text (Anxiety and computations of uncertainty during reward-based learning in volatile environments)
PSY_thesis_HeinT_2021.pdf - Accepted Version
Available under License Creative Commons Attribution Non-commercial No Derivatives.

Download (7MB) | Preview

Abstract or Description

Uncertainty plays a core mechanistic role in computational psychiatric accounts of clinical disorders. This thesis tests the hypothesis that subclinical anxiety interferes with uncertainty processing and reward learning in unstable environments. In a series of four experiments using a combination of computational modelling and reward-based learning tasks with electrophysiological and neuromagnetic data, we detail the relationships between anxiety, decision making, and uncertainty (inverse, precision) estimates. Using EEG and a hierarchical Bayesian filtering model of behaviour, the first experiment shows how experiencing a temporary state of anxiety can bias uncertainty estimates essential for optimal belief updates in Bayesian theories of learning and impede overall reward learning performance. We also tracked the expression of belief update signals in trial-by-trial EEG amplitudes, revealing precision-weighted prediction errors (pwPE) about stimulus outcomes were represented in the control group only. In the second study, we investigated how the behavioural and modelling findings from the first experiment are associated with the oscillatory representations of pwPEs and predictions hypothesised in predictive coding and Bayesian inference frameworks. Using convolution modelling for EEG oscillatory responses, we reveal the influence of biased uncertainty estimates in state anxiety on the oscillations encoding predictions and pwPEs during reward learning. For the third study, we asked how motivation to reduce anxiety may improve reward-based learning, detailing null results due to unsuccessfully inducing a state of anxiety. Using MEG, the final experimental chapter examined the effects of high levels of trait anxiety on reward learning and the spectral signatures of predictions and pwPEs. We show how high trait anxiety amplifies volatility estimates and increases uncertainty, impairing reward learning and altering the oscillatory responses encoding pwPEs. Together, the results from this thesis suggest that subclinical anxiety impedes reward-based learning by biasing uncertainty estimates and altering the neural encoding of pwPEs and predictions considered essential for optimal belief updating.

Item Type:

Thesis (Doctoral)

Identification Number (DOI):

https://doi.org/10.25602/GOLD.00030943

Keywords:

anxiety; uncertainty; hierarchical Bayesian inference; computational modelling; precision-weighted prediction error; predictive coding; oscillations; EEG; convolution

Departments, Centres and Research Units:

Psychology

Date:

30 November 2021

Item ID:

30943

Date Deposited:

21 Dec 2021 13:26

Last Modified:

18 Feb 2022 17:36

URI:

https://research.gold.ac.uk/id/eprint/30943

View statistics for this item...

Edit Record Edit Record (login required)