Late Integration of Prior Expectations During Precision-Weighted Perceptual Decisions

Adaptive perceptual decision-making relies on the ability to combine sensory evidence with prior expectations in a precision-weighted manner. Although Bayesian inference provides a clear normative account of how such integration should occur, the neural mechanisms through which the brain represents and combines priors and likelihoods remain poorly understood. Across two preregistered experiments, we investigated how the precision of learned priors and sensory likelihoods influence visual motion judgements and associated neural activity patterns during a random-dot motion estimation task. Neurotypical adult participants (N=80, 58 female) reported directions of visual motion stimuli, with motion coherence varying randomly across trials. Prior expectations over the continuous motion direction space were manipulated block-wise by implicitly varying the probabilities with which different motion directions were presented. Consistent with precision-weighted inference, response accuracy improved as motion coherence increased with robust response biases toward the expected motion direction. Neural activity measured using electroencephalography (EEG) revealed reliable effects of prior expectation on the univariate central-parietal positivity (CPP), consistent with reduced accumulation of sensory evidence under informative priors. Multivariate analysis using inverted-encoding models revealed robust effects of prior precision on stimulus-specific neural representations, but only late, during response planning stages. Together, these findings suggest that priors learned over a continuous space do not modulate the encoding of sensory evidence and are instead integrated later in the decision-process.Significance Statement Perceptual decisions require combining noisy sensory input with prior expectations about what is likely to occur. It remains unclear how the brain integrates prior expectations with incoming sensory evidence. Using a visual motion task and concurrent brain imaging, we show that expectations do not alter early stimulus-encoding, but are instead integrated at later stages of decision formation and action planning. These findings add nuance to influential theories proposing that learned expectations directly affect stimulus representations and instead highlight their contribution to later decision-related processes. This work advances understanding of how the brain uses experience to optimise perceptual decisions under uncertainty.