"Viewpoint Invariance in the Brain"
Primates are keen observers who are able to recognize a large variety of objects and faces across a range of viewing conditions. Underlying this capacity is a multi-area network of cortical and subcortical regions: the visual system. Distributed across large parts of the occipital, temporal and parietal lobes, a key task of the system is to achieve invariance, i.e. a robust representation of the outside world despite changes in illumination, size, position, and, perhaps most strikingly, changes in 3D viewpoint. While visual invariance has long fascinated researchers across many disciplines, we are only starting to understand which cortical mechanisms underlie this impressive feat. In this talk, I will focus on models of 3D viewpoint invariance with an emphasis on a potential representational shortcut: viewpoint symmetry. Predicted based on theoretical considerations regarding representational efficiency, viewpoint symmetry (i.e. joint selectivity for mirror-symmetric viewing angles) has recently been demonstrated through neurophysiological recordings in the macaque. I will present converging evidence from human participants (fMRI, EEG and TMS) suggesting that viewpoint symmetry is prevalent in the human visual system. These findings have implications for deep neural network architectures, and our understanding of reading acquisition in children.
Oct 31, 2016 | 04:00 PM