Spatio-temporal mechanisms of generative perception

2026-07-22

How do we rapidly and effortlessly compute a vivid veridical representation of the external world from the noisy and ambiguous input supplied by our sensors? One possibility is that the brain does not process all incoming sensory information anew, but actively generates a model of the world from past experience, and uses current sensory data to update that model. This classic idea has been well formulised within the modern framework of Generative Bayesian Inference. However, despite these recent theoretical and empirical advances, there is no definitive proof that generative mechanisms prevail in perception, and fundamental questions remain. The ambitious aim of GenPercept is to establish the importance of generative processes in perception, characterise quantitatively their functional role, and describe their underlying neural mechanisms. With innovative psychophysical and pupillometry techniques, it will show how past perceptual experience is exploited to manage and mould sensory analysis of the present. With ultra-high field imaging, it will identify the underlying neural mechanisms in early sensory cortex. With EEG and custom psychophysics it will show how generative predictive mechanisms mediate perceptual continuity at the time of saccadic eye movements, and explore the innovative idea that neural oscillations reflect reverberations in the propagation of generative prediction and error signals. Finally, it will look at individual differences, particularly in autistic perception, where generative mechanisms show interesting atypicalities. A full understanding of generative processes will lead to fundamental insights in understanding how we perceive and interact with the world, and how past perceptual experience influences what we perceive. The project is also of clinical relevance, as these systems are prone to dysfunction in several neuro-behavioural conditions, including autism spectrum disorder.

Italy

2026-02-01