Prediction error computation in visual perception

Prior knowledge plays a central role in shaping our perception and interactions with the world.

Predictive coding theories represent an influential framework formalizing the neural computations that underlie the integration of prior knowledge into perception.

One central tenet of predictive coding is the computation of prediction errors, which signify the discrepancy between predictions, derived from our prior knowledge, and incoming sensory input.

Despite ample empirical evidence supporting the existence of prediction errors, such as heightened sensory responses to unexpected compared to expected stimuli, several unresolved questions persist.One pivotal question pertains to the nature of the surprise encoded in sensory prediction errors across the visual cortical hierarchy.

Does the computation of prediction errors reflect local surprise, specific to each level in the visual hierarchy (e.g., surprising orientations in primary visual cortex)?

Alternatively, is the coding of surprise inherited top-down, aligning with the notion of predictive coding that predictions are relayed from higher to lower areas?

Furthermore, it remains unclear whether the computation of surprise dynamically adapts to our goals or remains a static feature of cortical tuning.

Lastly, the networks and temporal dynamics that underlie prediction and prediction error computations are poorly understood.Within the scope of the PreVision project, my objective is to address these critical knowledge gaps through the integration of advanced neuroimaging techniques and artificial intelligence methodologies, thereby shedding light on the fundamental neurocomputational mechanisms that drive perception.

Consequently, this project represents an excellent opportunity for me to engage in innovative and pioneering research, investigating pressing questions within contemporary cognitive neuroscience, and significantly advance my scientific career.