A neurocomputational investigation of human face processing

点击翻译按钮获取中文摘要

Face recognition plays a critical role in the ability to navigate our every-day physical and social environment. As such, it is the focus of significant research efforts in cognitive neuroscience. Overall, this research has contributed a lot to our knowledge regarding functional localization (i.e., what brain areas host face representations). In contrast though, much less is known about the actual nature of face representations (i.e., what type of neural code supports them). The overarching goal of the current proposal is to help elucidate this code along with the computations responsible for its generation. Three outstanding issues guide this investigation. First, what is the statistical structure of neural face representations as revealed by activation patterns in the human ventral cortex? Second, what encoding mechanism is responsible for mapping low-level image properties of face stimuli onto high-level neural face representations? And last, how can one invert this mechanism into reconstructing face images from the neural patterns that they elicit? To address these issues, the investigation combines human psychophysics, functional magnetic resonance imaging (MRI), and computational modeling along with novel statistical and machine learning techniques for the analysis of empirical data. In summary, the research proposed here aims to provide an algorithmic-level account of the relationship between pixel-based stimulus codes and neural-based representational codes in the domain of face perception. Ultimately, this work aims to shed light on what it means to recognize a face and on how our brains are able to accomplish this amazing perceptual feat.