Reinforcement learning biases in general and clinical population

Abstract: Reinforcement learning (RL) encompasses different applications in machine learning, psychology, and neuroscience, emerging from the combination of animal learning theories and mathematical algorithms. It is a great tool to study decision-making and explore inter-groups differences. In this work, we use RL modeling to investigate decision making biases that take place at different stages of decision-making and that are expressed in various ways depending on the population. We focus especially on (1) negativity bias in patients suffering from major depressive disorder but also (2) context dependence bias, and (3) positivity bias in visual outcome sampling in  the general population. Computational models help us understand these biases, providing mechanistic explanations and hypotheses for their causes. This thesis aims to pursue the understanding of inconsistencies in human decision-making and contribute to the debate on bounded rationality in both the general and clinical populations.
 

Directeur de thèse: Stefano Palminteri

Jury :
-Fabien Vinckier (président)
-Camilla Nord (rapportrice)
-David Sander (rapporteur)
-Charlotte Jacquemot (examinatrice)
-Julien Bastin (examinateur)