Title: What's in a prediction error?

Speaker: David Richter (Mind, Brain and Behaviour Research Centre (CIMCYC), University of Granada)

Location: In Person & Online via Zoom

Abstract: Predictive processing theories propose that the brain perceives the world by generating predictions and computing prediction errors, the mismatch between expected and observed input. Such prediction errors are thought to be a core neurocomputational mechanism underpinning perception. Yet, despite extensive evidence that the brain does compute sensory prediction errors, a critical question has remained largely unanswered: What does the visual brain actually predict? Is it low-level sensory details (e.g., edges and contrasts), high-level visual features (e.g. textures and object parts), semantic properties, or all the above? In my talk, I will present converging evidence from fMRI, EEG, and computational modelling showing that prediction errors across the visual hierarchy, including in early visual cortex, predominantly reflect high-level visual surprise. Moreover, these modulations emerge rapidly, within <200ms of stimulus onset, and persist irrespective of task demands. I will discuss how these high-level error signals may be broadcast from higher to lower visual areas, and what this may mean for accounts of hierarchical predictive inference. Taken together, these results suggest that the brain's predictive machinery is tuned to higher-order structure in the sensory input, prioritising what matters most for adaptive behaviour.

Bluesky: @davidrichter.bsky.social

Website: www.richter-neuroscience.com