Seeing is predicting: Testing a predictive coding account of visual perception involving saccadic eye movements

Description du projet

Suivre les mouvements oculaires, tester le codage prédictif

Nos yeux sont constamment en mouvement, analysant et interprétant des données visuelles complexes. Chaque seconde, nous effectuons environ trois saccades oculaires, mouvements rapides des yeux qui font se diriger la fovéa vers des objets d’intérêt visuel. Cela ne signifie pas que la perception visuelle est instable. C’est même tout le contraire. Le projet C-Pre, financé par l’UE, déterminera si le codage prédictif, qui implique la transmission de signaux de prédiction et d’erreur de prédiction à travers les réseaux cérébraux, peut expliquer cela. Le projet retracera le parcours de ces signaux afin de mesurer et d’enregistrer l’activité cérébrale et de déterminer des modèles expérimentaux de suivi oculaire. Les résultats obtenus feront progresser les sciences cognitives grâce à leur tentative très novatrice de falsification d’une théorie influente.

Objectif

We make about three saccadic eye movements per second which bring about an abrupt and drastic change in visual input on the retina. However, visual perception seems remarkably stable. I want to determine whether this phenomenon can be explained by predictive coding. Crucial for predictive coding is the transmission of prediction and prediction error signals across brain networks. I want trace down these signals in humans with non-invasive neurophysiological and neuroimaging methods. I will record EEG and MEG data combined with eye-tracking in gaze-contingent experimental designs in which visual input changes during a saccade and in which the frequency of change is manipulate to affect higher-level trans-saccadic expectations. I will apply measures of information transfer and dynamic causal modelling to assess brain dynamics time-locked to saccadic eye-movements. This approach will provide evidence for or against pre-saccadic top-down prediction and post-saccadic bottom-up prediction error signals, which will eventually speak to the question whether predictive coding can explain the phenomenon of visual stability. The proposed action will strongly enhance my theoretical knowledge about predictive processes and it will widely extend my methodological skills about MEG, source analysis, and the analysis of brain dynamics and connectivity. The project will be conducted under the supervision of Floris de Lange at the Donders Institute, the Netherlands, and includes a secondment with Karl Friston and Rick Adams at UCL, UK, as well as cooperation with Stefano Panzeri from IIT, Italy. In sum, this project combines methods and approaches from an international set of world-leading experts in an unprecedented way to test a hypothesis derived from the currently most comprehensive theory of brain function. Therefore, this project will make a highly-innovative falsification attempt of an influential theory which could mean a crucial step in advancing the cognitive sciences.