Computational modelling of decision uncertainty

  • Nadim Atiya

Student thesis: Doctoral Thesis


This PhD thesis contributes towards the development and analysis of a
neural circuit computational model of perceptual decision uncertainty and
change-of-mind behaviour. This highly-interdisciplinary work integrates
biological neural network modelling, cognitive psychology and neurophysiology
of metacognitive behaviour, and the mathematics of dynamical systems
theory. The thesis begins with a concise review of the various experimental
observations of decision uncertainty and change-of-mind. This is done
through an overview of the evolution of the experimental tasks that have
been developed over the years to understand decision uncertainty in humans
and animals. This is followed by an overview of the existing models of
decision uncertainty and change-of-mind, with a focus on cognitive models
(dynamical and probabilistic) and neural models. The thesis has led to
three original research contributions. In the first contribution, the first
cortical neural circuit model of decision uncertainty and change-of-mind
is introduced, effectively unifying the two fields of study. The proposed
model accounts for a variety of behavioural and neural signatures of decision
uncertainty and change-of-mind, while explaining the shared neural mechanism
that links both metacognitive features. In the second contribution,
more rigorous theoretical analyses of the model are presented. This is done
through systematic variation of key model parameters proposed in the first
contribution. Furthermore, the robustness of the model is highlighted, and
reward rate is investigated to identify the impact of various parameter
values on optimal performance. In the third contribution, changes-of-mind are investigated in situations when additional evidence is not available
after the initial decision, a type of situation that has been neglected by
the current theoretical and experimental accounts. Using an experimental
task, it is demonstrated that changes-of-mind can occur in the absence
of new post-decision evidence. Furthermore, using a reduced version of
the proposed neural circuit model, the neural mechanisms underlying such
changes-of-mind are uncovered. In particular, it is shown that changes-of mind
in the absence of new post-decision evidence are strongly linked to
elevated neural activity in the uncertainty-encoding population of the model,
consistent with recent neurophysiological evidence implicating higher order
networks in change-of-mind behaviour. Overall, the three contributions
shed light on the neural circuit dynamics underlying decision uncertainty
and change-of-mind behaviour, and offer a biologically-motivated theoretical
framework for future investigations.
Date of AwardOct 2019
Original languageEnglish
SupervisorKongfatt Wong-Lin (Supervisor) & Girijesh Prasad (Supervisor)


  • Decision-making
  • Change-of-mind
  • Computational neuroscience
  • Artificial Intelligence

Cite this