Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

Deep learning as an alternative to global optimization in diffusion model for conflict tasks

Stonkute, Solveiga 2019. Deep learning as an alternative to global optimization in diffusion model for conflict tasks. PhD Thesis, Cardiff University.
Item availability restricted.

[img]
Preview
PDF (PhD Thesis) - Accepted Post-Print Version
Download (4MB) | Preview
[img] PDF (Cardiff University Electronic Theses Publication Form) - Supplemental Material
Restricted to Repository staff only

Download (389kB)

Abstract

To apply mathematical models of decision making in psychological research, researchers need ways to extract model parameters from behavioural studies. The expansion of the drift diffusion model to con ict tasks (DMC) (Ulrich, Schroter, Leuthold, & Birngruber, 2015) resulted in the model being non-differentiable, which means that the parameters of DMC can only be estimated. The current methods for recovering parameters from DMC rely on comparing reaction time (RT) distributions. Such methods will struggle to recover all DMC parameters well due to the of the solution space of DMC, which means that some parameters can be confused with others when RT distributions are compared. Following that, five global optimization algorithms from different optimization families were compared to create a benchmark for parameter recovery from DMC. The results revealed that differential evolution outperformed the other four optimization algorithms in recovery of parameters from both distributions with high and low trial numbers. Even though differential evolution is capable of recovering parameters well, it is very expensive in computational time, which means that researchers who do not have access to vast computational resources cannot apply DMC in their research. Due to this, deep learning was investigated in application of parameter recovery from DMC. The results showed that deep learning recovered all model parameters exceptionally well from RT distributions with large trial numbers, and as well as differential evolution from RT distributions with low trial numbers, which allows application of deep learning models in deployment pipelines that take seconds rather than months. Finally, deep learning models were applied in several experimental studies investigating the effects of speed-accuracy trade-off (SAT) in response inhibition and perceptual decision making tasks, and how the performance relates between the tasks and over two different testing sessions, and demonstrated the effects of SAT on DMC parameters in different tasks.

Item Type: Thesis (PhD)
Date Type: Completion
Status: Unpublished
Schools: Psychology
Subjects: B Philosophy. Psychology. Religion > BF Psychology
Funders: ESRC
Date of First Compliant Deposit: 12 August 2019
Date of Acceptance: 12 August 2019
Last Modified: 12 Aug 2019 09:41
URI: http://orca-mwe.cf.ac.uk/id/eprint/124845

Actions (repository staff only)

Edit Item Edit Item

Downloads

Downloads per month over past year

View more statistics