bioRxiv2016
Understanding the integration of top-down and bottom-up signals is essential for the study of perception. Current accounts of predictive coding describe this in terms of interactions between state units encoding expectations or predictions, and error units encoding prediction error. However, direct neural evidence for such interactions has not been well established. To achieve this, we combined EEG methods that preferentially tag different levels in the visual hierarchy: Steady State Visual Evoked Potential (SSVEP at 10Hz, tracking bottom-up signals) and Semantic Wavelet-Induced Frequency Tagging (SWIFT at 1.3Hz tracking top-down signals). Importantly, we examined intermodulation components (IM, e.g., 11.3Hz) as a measure of integration between these signals. To examine the influence of expectation and predictions on the nature of such integration, we constructed 50-second movie streams and modulated expectation levels for upcoming stimuli by varying the proportion of images presented across trials. We found SWIFT, SSVEP and IM signals to differ in important ways. SSVEP was strongest over occipital electrodes and was not modified by certainty. Conversely, SWIFT signals were evident over temporo- and parieto-occipital areas and decreased as a function of increasing certainty levels. Finally, IMs were evident over occipital electrodes and increased as a function of certainty. These results link SSVEP, SWIFT and IM signals to sensory evidence, predictions, prediction errors and hypothesis-testing - the core elements of predictive coding. These findings provide neural evidence for the integration of top-down and bottom-up information in perception, opening new avenues to studying such interactions in perception while constraining neuronal models of predictive coding.\n\nSIGNIFICANCE STATEMENTThere is a growing understanding that both top-down and bottom-up signals underlie perception. But how do these signals interact? And how does this process depend on the signals probabilistic properties? Predictive coding theories of perception describe this in terms how well top-down predictions fit with bottom-up sensory input. Identifying neural markers for such signal integration is therefore essential for the study of perception and predictive coding theories in particular. The novel Hierarchical Frequency Tagging method simultaneously tags top-down and bottom-up signals in EEG recordings, while obtaining a measure for the level of integration between these signals. Our results suggest that top-down predictions indeed integrate with bottom-up signals in a manner that is modulated by the predictability of the sensory input.