Dorottya Hetenyi¹ (), Peter Kok¹; ¹Department of Imaging Neuroscience, UCL Queen Square Institute of Neurology, University College London, London, UK

Predictions about future events significantly influence how we process sensory information. However, the neural mechanisms that maintain these predictions in anticipation of incoming stimuli remain unclear. In previous work, we demonstrated that predicted shape representations exhibit oscillatory activity in the alpha band (10 – 11 Hz) throughout pre-stimulus windows. In that work, participants’ task was orthogonal to the shape predictions. Here, we extended these findings by having participants perform a shape identification task that was directly related to the shape predictions, to allow us to link the neural correlates of prediction to behaviour. We used magnetoencephalography (MEG) combined with multivariate decoding techniques to examine the content and frequency characteristics of perceptual predictions. Participants performed a shape identification task in which auditory cues predicted which shape would likely appear (75% valid). To make the identification of the shapes challenging, they were embedded in white noise. First, we found that valid prediction cues improved both identification accuracy and reaction times. Signal detection theory analyses revealed that participants were significantly biased towards reported the predicted shape (i.e., reduced criterion) without affecting sensitivity (i.e., similar d-prime). Additionally, frequency analysis of MEG shape decoding signals revealed the oscillatory nature of predicted shape representations, specifically in the alpha band (10–11 Hz) during the pre-stimulus period. Together, these findings suggest that sensory predictions are represented in pre-stimulus alpha oscillations, and the brain may leverage alpha oscillations as a neural mechanism to balance prediction and perception.