Author: Law, Chun Kit
Title: Neural mechanisms underlying value computation
Advisors: Chau, Bolton (RS)
Degree: Ph.D.
Year: 2022
Award: FHSS Faculty Distinguished Thesis Award (2021/22)
Subject: Decision making -- Data processing
Choice (Psychology)
Neural networks (Neurobiology)
Hong Kong Polytechnic University -- Dissertations
Department: Department of Rehabilitation Sciences
Pages: v, 150 pages : color illustrations
Language: English
Abstract: It is not hard to imagine making a choice between items of the same category, for example, two watches vary in price, accuracy, design, etc. A decision can be made by comparing these attributes. This kind of everyday decisions has been extensively investigated over the past decades and the underlying neural mechanisms were reviewed in Chapter 1. However, it is also ubiquitous to face decisions between ostensibly incomparable options (e.g. wages versus vacation, staying in the current company versus searching alternatives in the job market). A question arises, how could these heterogeneous options be compared? Since there are countless decisions, it is impossible for the brain to have individual mechanisms for every decision. I therefore propose the brain should respond to decisions by the classes of the options involved. I introduce an approach of classification according to the qualitative differences between the options. As a result, options are categorized into either items or environments. Items are concrete options that provide direct payoff whereas environments are options which lead to potential impacts in the future as opposed to direct payoff (e.g. searching in the job market provides an opportunity to get better jobs but not directly provide an offer). With this classification, I combined human behavioural testing, deep learning neural network, and brain imaging to address the central question of this thesis - whether there is a neural mechanism which can make comparison among all different kinds of option or there are different neural mechanisms to work with particular decisions.
To address the central question of this thesis, first it is necessary to investigate environment choice (i.e. decision between environments), which receives little attention and remains highly elusive. In Chapter 2, I examined human decision-making specifically when decisions are made between environments via a behavioural study. In a binary decision-making task, environments were characterized by their complex structures - each environment was composed of 20 items and hence massive information was embedded. Yet, the results demonstrated that people were capable of integrating the information in the environments to guide their choices. Particularly, I found that people preferred environments with larger means or variances in their component item distributions. After that, the second study involved a two-stage decision-making task to directly contrast environment choice and item choice. Behavioural results reported in Chapter 3 showed that context-dependent adaptation, an essential property of decision-making, was observed in both environment choice and item choice. It served as the basis for developing computational models in Chapter 4 and examination of neural signals in Chapter 5.
To understand the neurocomputation during decision-making, in Chapter 4, I developed and tested different computational models to describe participants' choice behaviour in the studies reported in Chapters 2 and 3. Specifically, I employed convolutional neural network (CNN), general linear model, cumulative prospect theory, mean-variance-skewness model, power law model, and autoencoder to fit the behavioural data of those studies. Model comparison results showed that the CNN, a deep learning neural network, best describes participants' decision-making behaviour in both studies. The CNN possesses a strength over traditional computational models that it allows fewer a priori assumptions, by which implicit features of the valuation process can be captured and described even some they are not specified explicitly. Besides, it possesses multiple nodes and multiple layers for representation of option value, facilitating the examination of multivariate neural signals in Chapter 5. A closer inspection using a series of representational similarity analysis (RSA), a multivariate analysis, ascertained that the CNN multi-nodal representations encode the complex information of the item distributions in the environments.
Chapter 5 reports a functional magnetic resonance imaging (fMRI) study that showed a double dissociation of the lateral frontopolar cortex (FPl) and ventromedial prefrontal cortex (vmPFC) in environment choice and item choice. Closer inspection revealed both the FPl and vmPFC signals exhibited essential properties of value comparison process (e.g. activity correlated with the difference in value between options; invariance to salience), suggesting the FPl and vmPFC did subserve the value comparison process. In addition, I performed a series of RSA to test the similarity of the FPl multi-voxel activation patterns during environment choice with the multi-nodal representations of CNN variants that employ single or multiple feature detectors for encoding environment value. Notably, the FPl was found similar to the CNN with multiple feature detectors only, and the FPl was the most similar to the CNN developed in Chapter 4. It implies the FPl and CNN shared similar multiple parallel encoding processes in valuation of environments and the FPl carried a multivariate coding for the complex environment information.
To conclude, in this thesis I proposed an approach of classification to categorize options into items or environments according to their qualitative differences. With the use of behavioural testing, computational models, and brain imaging, I demonstrated environment choice and item choice were indeed dissimilar and they involved distinct neural mechanisms. Dissociable roles of the FPl and vmPFC in environment choice and item choice were revealed. It reflects a functional specialization in decision-making that the brain requires multiple neural substrates to work with particular types of decision as opposed to a single neural substrate to deal with all kinds of decision.
Rights: All rights reserved
Access: open access

Files in This Item:
File Description SizeFormat 
6200.pdfFor All Users3.55 MBAdobe PDFView/Open

Copyright Undertaking

As a bona fide Library user, I declare that:

  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.

By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.

Show full item record

Please use this identifier to cite or link to this item: