The neural dynamics of hierarchical Bayesian inference in multisensory perception

Abstract

Transforming the barrage of sensory signals into a coherent multisensory percept relies on solving the binding problem - deciding whether signals come from a common cause and should be integrated, or instead be segregated. Human observers typically arbitrate between integration and segregation consistent with Bayesian Causal Inference, but the neural mechanisms remain poorly understood. We presented observers with audiovisual sequences that varied in the number of flashes and beeps. Combining Bayesian modelling and EEG representational similarity analyses, we show that the brain initially represents the number of flashes and beeps and their numeric disparity mainly independently. Later, it computes them by averaging the forced-fusion and segregation estimates weighted by the probabilities of common and independent cause models (i.e. model averaging). Crucially, prestimulus oscillatory alpha power and phase correlate with observers prior beliefs about the worlds causal structure that guide their arbitration between sensory integration and segregation.