Testing the predictive coding in visual perception using Multivariate EEG analyses

Poster Session F - Tuesday, April 16, 2024, 8:00 – 10:00 am EDT, Sheraton Hall ABC

Olga Bulatova^1,2, Frida Printzlau^1,2, Adam Malitek^1,2, Keisuke Fukuda^1,2; ¹University of Toronto, ²University of Toronto Mississauga

The predictive coding theory of the brain posits that our brain generates predictions about upcoming percepts to facilitate sensory processing. However, the underlying neural mechanism remains unclear. One possibility is that the predicted percept is instantiated by anticipatory neural activity in the sensory regions. Alternatively, it might be instantiated through activity-silent mechanisms. To test this, we used multivariate decoding analyses to examine whether the predicted percept could be decoded from EEG activity prior to the onset of the predicted percept or from the activity generated by a visual “ping” (Wolff et al., 2017). Specifically, participants saw a picture prime (e.g., chair or a glass of drink; picture prime) followed by a letter-by-letter presentation of a three-letter word (e.g., s-i-t or s-i-p) that only differed in the last letter. Participants’ task was to anticipate the upcoming letter based on the picture prime and judge whether the word spelled out matched semantically with the picture prime. The picture prime and the word matched semantically 80% of the time. In addition, a large high-contrast patch (perceptual ping) appeared immediately preceding the presentation of the last letter in 20% of the trials. Preliminary results (n=16) suggest that the last letter can be decoded reliably above chance from the pattern of EEG activity among the parieto-occipital electrodes prior to the letter onset but not from the activities elicited by the ping. Overall, our preliminary results suggest that the predictive coding of upcoming visual percepts relies on active neural signals rather than activity-silent signals.

Topic Area: PERCEPTION & ACTION: Vision