Between prediction and reality: top-down propagation, communication and modulation

Expectations predict the upcoming visual information, facilitating its disambiguation from the noisy input to speed up behavior. However, the neural mechanisms that support such dynamic feature flow and how they facilitate behavior remain unclear. In this thesis, I will trace the propagation, communication and modulation effect related to the prediction. In the first study (Chapter 2), I initiated a cueing-categorization design and validated its feasibility. I first detected the sensibility of participants in distinguishing auditory pitches (i.e., the cues) by estimating their d-primes. Next, in two phases, I instructed participants to build the coupling relationship between auditory cues and stimuli, then proved that compared with non-informative prediction, informative ones can significantly reduce the reaction time. Recognizing confounding factors in the first study, I further improved the design by manipulating two specific and separate predicted contents. I used a prediction experiment that cued participants (N = 11) to the spatial location (left vs. right) and spatial frequency (SF, Low, LSF, vs. High, HSF) contents of an upcoming Gabor patch. I reconstructed two networks (prediction network and categorization network) in the following two studies with simultaneous MEG recordings of each participant’s neural activity. In the second study (Chapter 3), focusing on the pre-stimulus prediction stage, I answered when, where and how predictions dynamically propagate through a network of brain regions. I traced the dynamic neural representation of predictive cues and reconstructed the communications about predicted contents in a functional network, sequentially from temporal lobe at 90-120ms, to occipital cortex after 200ms, with modulatory supervision of frontal regions at 120-200ms. In the third study (Chapter 4), turning to the post-stimulus stage, I reconstructed the communication network propagating the stimulus feature from occipital-ventral regions (150-250ms) to parietal lobe (250-350ms), finally arriving premotor cortex (>350ms) which modulates behavioral categorization. I found the previous prediction previewed and then sharpen stimulus representation across the categorization network, leading to a faster reaction time. I discussed the generalization of the findings to other stimulus features and sensory modalities. Putting forward the plans about developing a series of structured studies on predicting higher-dimensional features, in the future, I aim to understand the neural mechanisms about how prediction tunes perception and to trace the concrete predicted contents in laminar layers with the fusion of E/MEG and fMRI