By Yves Lherry In On April 6, 2017
Learning by seeing or doing? Assessing the overlap between observational and instrumental inference during decision-making: computations, neural correlates and alterations in psychosis
Making decisions, from perceptual judgments to policy-making orientations, often requires to combine multiple pieces of ambiguous or conflicting information. In such conditions, human choices exhibit a suboptimal variability whose origin and structure remains poorly understood. Dominant psychological theories attribute this choice variability to a mixture of cognitive biases and random noise in an otherwise optimal decision process. However, this ‘normative’ approach falls short of explaining the core variability of human decisions and its implications in terms of cognitive and neural architecture. This PhD project proposes to take a rather different perspective, by seeking to understand the suboptimality of human decision-making from its underlying neurobiology. We hypothesize that human excursions from optimality reflect general neural coding constraints imposed by the functional architecture of inference in the brain. To test this hypothesis, we plan to identify neural coding constraints using non-invasive measures of electromagnetic brain activity in magnetoencephalography (MEG), the only neuroimaging technique combining high temporal resolution (of the order of 1 ms) with whole-brain coverage at a relatively focal spatial resolution (of the order of 1 cm). The recorded data will be analyzed using an original methodological framework which combines state-of-the-art multivariate pattern analysis techniques particularly suited to such signals with a computational model of noisy statistical inference.
This PhD project will test the generality and specificity of identified neural coding constraints across two separate, almost non-communicating topics of research in decision sciences: perceptual categorization and reward-guided learning. Past work has identified, through different paradigms and questions, spatially dissociable neural circuits for perceptual and reward-guided decisions. We will design an original paradigm in which the same stimuli can correspond, in different conditions, either to cues about a hidden state of the environment that subjects have to guess, or to outcomes of their previous choice. We hypothesize that the human brain has evolved with partially dissociable brain networks in link with cue- and outcome-based inference, and that the different architectural properties of these two brain networks can explain selective differences in decision-making and pathologies of decision-making.
A first MEG study in healthy subjects will assess the degree of spatial overlap between the neural coding constraints underlying cue- and outcome-based inference in this paradigm (year 1). Two clinical EEG studies (years 2-3) will then test selective pathologies of outcome-based inference in two clinical conditions leading to psychosis: schizophrenia and 22q11.2 gene deletion syndrome. We will test whether psychotic patients are selectively impaired in their ability to track action-outcome associations, caused by an overexpressed influence of uncertainty about the relationship between the patient’s action and its outcome. This PhD project will be conducted in close collaboration with Pr. Karim Jerbi at the University of Montréal, a leading specialist in the machine learning methods which we plan to apply to analyze the recorded MEG and EEG data. For the two clinical studies, we will collaborate with psychiatrists with long-term experience in the study of clinical populations suffering from psychosis at the Centre Hospitalier Le Vinatier in Lyon (Dr. Nicolas Franck and Dr. Caroline Demily).
Name : Inserm U960
Director : Etienne KOECHLIN
Supervisor(s) : Valentin WYART
Team leader : Etienne KOECHLIN
Funding : FdV
e-mail : email@example.com