Pierre Costa¹ (), Thérèse Collins¹; ¹Integrative Neuroscience and Cognition Center, Université Paris Cité & Centre National de la Recherche Scientifique (CNRS), Paris, France

Our environment is constantly changing but our perception is stable. This stability may result from the integration of successive visual images that smooths over spurious variations in input. Such a process manifests in a behavioral phenomenon called Serial Dependence (SD), an attractive bias between successive stimuli. Here we questioned whether SD operates on low-level stimulus representations or at a higher level. We constructed face morphs from three prototypes of the same face expressing anger, fear and sadness. We measured the perceived similarity between each pair of morphs with an odd-one-out experiment, to build a behavioral representational matrix. As expected, emotional expressions were perceived categorically, with greater perceived similarity between pairs within a category than between pairs across categories, for a fixed distance in stimulus space. Subjects then reproduced the perceived facial expression of successively presented morphs by adjusting a response cue while their cerebral activity was recorded with EEG. The categorical effect was mirrored in the EEG signal in a given time window and group of electrodes. Using representational similarity analysis, we computed an EEG-based correlation matrix based on the similarity between each pair of morphs. This neural representational matrix correlated more with the behavioral representational matrix than with the stimulus matrix. We also correlated the three matrices (stimulus, behavioral, neural) with the SD matrix, i.e. the extent to which perceptual reports were biased towards the facial expression seen in the past (SD). SD was stronger between two faces belonging to the same category than between two faces that did not belong to the same category, for a given stimulus difference. The SD matrix correlated better with the representational matrices than the stimulus matrix. These results reveal categorical SD, showing that the temporal integration serving visual stability operates on high-level visual representations.

Acknowledgements: ANR CNRS