Skip to Main content Skip to Navigation

Apprentissage et contrôle cognitif : une théorie computationnelle de la fonction exécutive préfontale humaine

Abstract : Cognitive control enables appropriate action selection according to stimuli, but also present context or past cues, while taking our internal goals into account. Cognitive control and learning are profoundly and reciprocally linked. On one side, cognitive control requires that a repertoire of behaviors be learnt, as well as their values in different conditions, for appropriate use. On the other side, cognitive control is needed for learning of a repertoire of behaviors, notably to regulate the exploration-exploitation trade-off, but also to generalize, decide to switch, infer a structure in a problem, etc. . ..Thus, cognitive control and learning are strongly linked in the flexibility that characterizes human prefrontal executive function. However, this link is presently poorly understood and few psychological or cognitive neuroscience studies include both aspects. Moreover, existing computational models of learning and decision do not account for their interaction. In this PhD thesis, we propose a mathematical theory combining reinforcement learning and Bayesian inference mechanisms. This model includes learning of repertoires of behaviors (task-sets) in an uncertain environment as well as cognitive control (task-switching) in presence or absence of contextual information. This model makes specific predictions that we tested in two behavioral experiments. They validate the predictions of the theory against other existing models. Moreover, the theory proposes an explanatory factor for qualitative differences in exploratory strategies that we observed across individuals. The proposed theory intrinsically characterizes essential notions such as default behavior, switch and exploration. It allows for the natural emergence of a control mechanism of the exploitationexploration trade-off, as well as its weighing factor. Lastly, empirical results validate the predictions and confirm the hypotheses of the model. The model may be used in functional imaging studies to understand computations executed in the brain, with prefrontal cortex, basal ganglia and neurotransmitters such as dopamine and norepinephrine as main points of interest.
Document type :
Complete list of metadatas

Cited literature [209 references]  Display  Hide  Download
Contributor : Theses Bupmc <>
Submitted on : Wednesday, April 17, 2013 - 5:13:01 PM
Last modification on : Thursday, December 10, 2020 - 12:38:41 PM
Long-term archiving on: : Thursday, July 18, 2013 - 4:09:07 AM


  • HAL Id : tel-00814840, version 1


Anne Collins. Apprentissage et contrôle cognitif : une théorie computationnelle de la fonction exécutive préfontale humaine. Neurosciences. Université Pierre et Marie Curie - Paris VI, 2010. Français. ⟨NNT : 2010PA066124⟩. ⟨tel-00814840⟩



Record views


Files downloads