Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.742554
Title: Reinforcement learning approaches to the analysis of the emergence of goal-directed behaviour
Author: Mitsopoulos, Constantinos
ISNI:       0000 0004 7230 015X
Awarding Body: Birkbeck, University of London
Current Institution: Birkbeck (University of London)
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Over recent decades, theoretical neuroscience, helped by computational methods such as Reinforcement Learning (RL), has provided detailed descriptions of the psychology and neurobiology of decision-making. RL has provided many insights into the mechanisms underlying decision-making processes from neuronal to behavioral levels. In this work, we attempt to demonstrate the effectiveness of RL methods in explaining behavior in a normative setting through three main case studies. Evidence from literature shows that, apart from the commonly discussed cognitive search process, that governs the solution procedure of a planning task, there is an online perceptual process that directs the action selection towards moves that appear more ‘natural’ at a given configuration of a task. These two processes can be partially dissociated through developmental studies, with perceptual processes apparently more dominant in the planning of younger children, prior to the maturation of executive functions required for the control of search. Therefore, we present a formalization of planning processes to account for perceptual features of the task, and relate it to human data. Although young children are able to demonstrate their preferences by using physical actions, infants are restricted because of their as-yet-undeveloped motor skills. Eye-tracking methods have been employed to tackle this difficulty. Exploring different model-free RL algorithms and their possible cognitive realizations in decision making, in a second case study, we demonstrate behavioral signatures of decision making processes in eye-movement data and provide a potential framework for integrating eye-movement patterns with behavioral patterns. Finally, in a third project we examine how uncertainty in choices might guide exploration in 10-year-olds, using an abstract RL-based mathematical model. Throughout, aspects of action selection are seen as emerging from the RL computational framework. We, thus, conclude that computational descriptions of the developing decision making functions provide one plausible avenue by which to normatively characterize and define the functions that control action selection.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.742554  DOI: Not available
Share: