A recurrent neuronal model for the effect of predictions on sensory processes

Abstract

The effects of prior knowledge, predictions, and expectations on sensory and decision-making processes have been extensively studied. Yet, the neural mechanisms underlying those effects are still unclear. Here, we propose a recurrent neuronal model and test its predictions on behavioral and neuroimaging data from the literature. The model implements predictive processing through recurrent interactions among feature-tuned neural populations that integrate feedback and feedforward signals within established cortical circuitry. Our results show that the model can successfully explain the behavioral effects of prediction found in a previous study in which houses and faces were used as stimuli. We then simulate fMRI data using the protocols of three different studies from the literature, and the optimized model parameters obtained from the model fit to the behavioral data. Although the studies used diverse visual stimuli, not limited to faces and houses, the model predicts their findings to a great extent, proving its generalizability. Overall, our findings demonstrate that the proposed model can provide a link between behavior and neural activity, offering a theoretical account of how prior knowledge, predictions, and expectations influence sensory processing.