This week's NCA group meeting presents: Tibi Tesileanu who will be presenting and his talk regarding: Biological learning of local motion detectors! Please see the detailed description below.
Motion detection is a fundamental task for the visual system, with cells as early as the retina showing selectivity to specific directions of motion. These cells typically have localized connections and receptive fields. We provide a normative model for the connectivity pattern in motion-sensitive cells based on the assumption that they are adapted to typical motions encountered in natural environments. We model localized motions as rotations in the high-dimensional pixel space, which allows us to use concepts from the theory of continuous groups. We show that, when the training data contains naturalistic patterns undergoing localized motion, a sparse-coding approach learns receptive fields involving small sets of nearby pixels. We implement the sparse-coding step in our model using non-negative similarity matching, leading to a biologically-plausible algorithm.