QLS Seminar Series - Angela Yu

Computational Modeling of Human Face Processing

Angela Yu, Technical University of Darmstadt
Tuesday December 5, 12-1pm
Zoom Link: https://mcgill.zoom.us/j/86855481591
In Person: 550 Sherbrooke, Room 189

Abstract: Face processing plays a central role in everyday human life, and provides an ideal case study of a complex and realistic, yet focused and interpretable domain of computational investigation of cognitive processes. My group has investigated the computational nature of face representation and processing in the brain, by integrating and developing a variety of machine learning and computer vision methods. We have found that human social trait perception has both a linear component and a quadratic component, with the latter specifically related to the statistical typicality of a face. We have shown that this typicality element is related to efficient coding in feature representation, with implications for learning and exploration. Moreover, we find task-dependent attention modifies face feature representation by re-representing faces in the most task-relevant feature subspace, with predictable consequences on corresponding typicality/attractiveness judgement. Separately, we estimate the relative importance humans assign to different features of the face, and show that these correspond well to the facial features that best objectively discriminate individuals, with the caveat that these attentional weights are modified by personal experience with different types of faces. Finally, I will discuss how facial processing affects social perception and judgment, with implications for gender and racial biases.