UC Santa Barbara

Face recognition is a ubiquitous task that humans are adept at. Recent research has estimated that adult humans typically view faces for about 20% of every waking hour (Oruc et al., 2019). Although the behavioral and neural signatures of face processing have been well studied, little is known about how long-term visual experience shapes them. The incoming visual experience of faces is thought to have a changeable component (such as expression and gaze direction) and an invariant component (such as the configuration of features, see Haxby et al., 2000). One claim that is often made is that the aspects of the visual experience that are invariant influence the underlying neural computations and, consequently, the perception of faces(Diamond & Carey, 1986). However, we do not fully understand the scope of this influence and the computational mechanisms by which this influence is exerted. This thesis aims to address this knowledge gap. Specifically, we examined two well-established kinds of invariance in our visual experience of faces: one arising from the prolonged use of a consistent oculomotor strategy to view faces (Peterson & Eckstein, 2012), and the second arising from a preponderance of viewing faces in an upright configuration (Yin, 1969). Earlier research has shown that humans land their first fixation around a consistent point on the face, known as the preferred fixation location (PFL). Human performance in various common perceptual tasks like person-identification, gender categorization, and emotion recognition are shown to be tuned to the PFL (Peterson & Eckstein, 2012). However, not known is whether other perceptual effects in faces are also tuned to the PFL. Chapter I studied how the long-term oculomotor strategy of moving the eyes to the PFL shapes two well-studied perceptual effects: the Composite Face effect (CFE, Young, et al., 1987) and the Face Adaptation Effect (FAE, Webster, et al., 2004). For this, we compared the strengths of these effects in two groups of observers: one with a PFL close to the eyes (upper lookers) and the other with a PFL lower on the face around the nose region (lower lookers). We found that the PFL modulates both effects. On the one hand, the CFE was smaller for the lower lookers. On the other hand, the FAE was more fixation position-specific for observers whose PFL was farther away from the eyes. These findings extend the scope of the influence of visual experience on face perception to the CFE and the FAE. Humans view upright faces disproportionately more often than inverted faces. The perceptual consequence of this predilection is the well-known face inversion effect(FIE, Yin, 1969). Tsank (2019) related the FIE to visual experience using a Convolutional Neural Network (CNN). However, the mechanism by which the tuning to upright faces as a result of visual experience to upright but not inverted faces develops is unknown. Recent research has suggested that the face inversion effect arises due to spatial summation in higher-level visual regions in the brain (Poltoratski et al., 2021). Therefore, we hypothesized that visual experience influences the spatial summation ability for upright and inverted faces. To study this, we used a spatial summation task where observers had to identify upright or faces covered by apertures of different sizes. Our results revealed that as long as the apertures are not too small, humans’ spatial summation efficiency increases with aperture size for upright faces but reduces for inverted faces. We then used a convolutional neural network (CNN) that was trained on full upright or inverted faces to show that the network developed superior spatial integration abilities only for the orientation of faces it was trained on. Moreover, the divergence in spatial integration efficiency for upright vs. inverted faces only appears when a network with a large receptive field is used, supporting the hypothesis that higher visual areas in the brain mediate the FIE. Our results extend our understanding of the role of prolonged use of a consistent oculomotor strategy to view faces in shaping perceptual effects in faces. We also illustrate the utility of model observers in constraining hypotheses about computational mechanisms driving perceptual effects. Together, this research furthers our understanding of the unique perceptual and computational consequences of visual experience to faces.