| Project Detail |
The brain uses prior experiences to make a “best guess” of the noisy and ambiguous information in the world around us. These predictions are constantly updated as new sensory information becomes available, yet the computational resources of the brain are limited. What is a viable model for representing and updating predictions in the brain that is both computationally efficient and neurobiologically plausible? I hypothesise that the brain represents predictions by decomposing them into their constituting features and pre-activating those features considered instrumental in explaining the sensory input, thereby allowing to update only parts of the prediction without revising the whole prediction. The proposed project fuses concepts of multi-dimensional feature encoding, cortical processing hierarchies and activity-silent working memory to develop a computational model of prediction and error coding using artificial neural networks. The implications of this model will be empirically tested in human visual perception using functional magnetic resonance imaging with laminar precision. This project aims to gain a deeper understanding of the compositionality of predictions as a building block of fast and accurate perception and bears the potential to transform our understanding of the neural mechanisms giving rise to predictive processes. The project will be carried out at the Donders Institute for Brain, Cognition and Behaviour in the lab of Floris de Lange, a leading expert in the field of predictive processing. As an MSCA postdoctoral fellow, I will conduct cutting-edge research at the intersection of psychology, neuroscience and artificial intelligence, which will give me the opportunity to hone my methodological and management skills, as well as develop my own research program. |
