2018 Conference on Cognitive Computational Neuroscience

5-8 September 2018    Philadelphia, Pennsylvania

Technical Program

Paper Detail

Paper: PS-1A.9
Session: Poster Session 1A
Location: Symphony/Overture
Session Time: Thursday, September 6, 16:30 - 18:30
Presentation Time:Thursday, September 6, 16:30 - 18:30
Presentation: Poster
Publication: 2018 Conference on Cognitive Computational Neuroscience, 5-8 September 2018, Philadelphia, Pennsylvania
Paper Title: Understanding illusory face perception in the human brain
Manuscript:  Click here to view manuscript
DOI: https://doi.org/10.32470/CCN.2018.1203-0
Authors: Susan Wardle, Jessica Taubert, National Institutes of Health, United States; Lina Teichmann, Macquarie University, Australia; Chris Baker, National Institutes of Health, United States
Abstract: Face pareidolia is the spontaneous misperception of illusory faces in inanimate objects. As a natural error of face detection, investigation of face pareidolia has potential to reveal new insight into the mechanisms underlying face and object recognition in the primate brain. To understand the temporal dynamics of illusory face perception, here we used magnetoencephalography (MEG) to measure the brain activation patterns of 22 human participants in response to 32 illusory faces in inanimate objects, 32 similar non-face objects, and 32 human faces. We compared the time-varying brain activation patterns to the representations of the stimuli obtained with common pre-trained deep neural networks and computational models of visual saliency. Whereas the saliency models showed an early and limited correlation with the MEG data peaking at ~100ms after stimulus onset, the deep neural networks showed a greater and sustained correlation across the time-course which generally increased with layer depth. However, both forms of computational model were outperformed by a simple model which grouped stimuli by category (faces versus objects). Together the results highlight that the dynamic response of the human brain to complex natural stimuli is only partially captured by existing computational models.