Temporal dynamics of the visual representation of orientation ensemble perception

Humans can extract summary statistics of multiple visual elements. Numerous behavioral experiments have shown that ensemble perception occurs rapidly and accurately across several visual features (e.g., orientation, size, face). The visual system needs to incorporate individual elements efficiently to achieve such rapid perception, but it has yet to be clear how visual representation evolves over time during ensemble perception. The present study addressed this question by exploring what information (e.g., individual elements or their ensemble) can be decoded from spatiotemporal patterns of brain signals after presentation of multiple visual stimuli. Specifically, we measured EEG signals in two experiments where six Gabor patterns were presented in a circle. To construct decoders predicting orientation from EEG patterns, we first conducted an orientation discrimination task where human observers judged whether Gabor patterns were tilted clockwise or counterclockwise relative to the vertical. In this experiment, all Gabor patterns had the same orientation (from 15 to 165 degrees in steps of 30 degrees). We constructed orientation decoders for each time point and confirmed that decoding accuracy reached significance at around 100ms after stimulus onset and remained above chance for several hundred milliseconds. Next, we applied these decoders to EEG signals in a second experiment where observers performed an ensemble perception task for a set of six Gabor patterns with different orientations. Observers indicated whether average orientation of the set was tilted clockwise or counterclockwise relative to the vertical. To track the change in visual representation during the task, we calculated posterior probabilities of orientations in the set by using outputs of the decoders. We found that representation of individual elements with large orientation, rather than average orientation of the set, was dominant until 300ms after stimulus onset. These results suggest that ensemble processing might occur more sluggishly than expected from previous studies.