In Search of Invariance in Brains and Machines

Abstract:

Despite their seemingly impressive performance at image recognition and other perceptual tasks, deep convolutional neural networks are prone to be easily fooled, sensitive to adversarial attack, and have trouble generalizing to data outside the training domain that arise from everyday interactions with the real world. The premise of this talk is that these shortcomings stem from the lack of an appropriate mathematical framework for posing the problems at the core of deep learning - in particular, modeling hierarchical structure, and the ability to describe transformations, such as variations in pose, that occur when viewing objects in the real world. Here I will describe an approach that draws from a well-developed branch of mathematics for representing and computing these transformations: Lie theory.  In particular, I shall describe a method for learning shapes and their transformations from images in an unsupervised manner using Lie Group Sparse Coding.  Additionally, I will show how the generalized bispectrum can potentially be used to learn invariant representations that are complete and impossible to fool.

Bio:

Bruno OIshausen is Professor of Neuroscience and Optometry at the University of California, Berkeley.  He also serves as Director of the Redwood Center for Theoretical Neuroscience, an interdisciplinary research group focusing on mathematical and computational models of brain function.  He received B.S. and M.S. degrees in Electrical Engineering from Stanford University, and a Ph.D. in Computation and Neural Systems from the California Institute of Technology.  Prior to Berkeley he was a member of the Departments of Psychology and Neurobiology, Physiology & Behavior at UC Davis.  During postdoctoral work with David Field at Cornell he developed the sparse coding model of visual cortex which provides a linking principle between natural scene statistics and the response properties of visual neurons.  Olshausen's current research aims to understand the information processing strategies employed by the brain for doing tasks such as object recognition and scene analysis.  This work seeks not only to advance our understanding of the brain, but also to discover new algorithms for scene analysis based on how brains work.