[4T1-OS-6c] OS-6

[4T1-OS-6c-01] Exploring the Similarities in Face Recognition between Deep Learning and Individuals with Autism Spectrum Disorders: An Example using FaceNet

〇Taku Imaizumi¹, Lu Li¹, Natsuki Nishikawa², Hirokazu Kumazaki², Kazuhiro Ueda³ (1. Graduate School of Interdisciplinary Information Studies, The University of Tokyo, 2. Graduate School of Biomedical Sciences, Nagasaki University, 3. Graduate School of Arts and Sciences, The University of Tokyo)

Keywords:FaceNet, face recognition, uncanny valley, autism spectrum disorder

Although deep learning-based face recognition is highly accurate, it often diverges from human-like judgments. Notably, it is known that FaceNet failed to replicate certain features of the uncanny valley due to the large discrepancy between the human evaluation and the FaceNet evaluation for specific face images. For the images that FaceNet rated highly human-like, localized attention, such as the mouth and chin, acted as the basis for judgment. Such localized attention is consistent with the characteristics of people with autism spectrum disorder (ASD). This study investigated the similarity between FaceNet ratings and those of individuals with ASD. Regression analyses were conducted with FaceNet ratings as the dependent variable and ratings by typically developing individuals (TD)/those with ASD as independent variables. The findings revealed a closer resemblance between FaceNet ratings and those of ASD individuals.