| Description |
Natural images are scale invariant, with structures that span a vast range of length scales – from the smallest within a few dozen pixels to large structures the size of the image itself. In this talk, I will present a novel statistical mechanical model of natural images rooted in the theory of critical phenomena that sheds light on this hierarchy of length scales, paving the road for a (nonequilibrium) statistical model of natural images with few parameters. Studies of natural image statistics were first done in neuroscience but have since become of immense interest in machine learning due to the recent advances in deep learning. My research on natural images fits into a bigger picture: the general and common problem of "many scales of length" in physics, machine learning and neural coding. |
Problems in physics, machine learning, and computational neuroscience with many scales of length
Go to day