Normalization in Deep Convolutional Neural Networks

DATE/TIME CHANGE: May 21st (6pm) at Moonlighter

Research talk by Luis Gonzalo Sanchez Giraldo (http://web.cs.miami.edu/home/lgsanchez/)

Full title: “Normalization in Deep Convolutional Neural Networks: Incorporating nonlinearities inspired by the brain”

Abstract: Deep neural networks, which have now become the state of the art in object and scene recognition, have also shown an intriguing ability to explain some aspects of visual neural processing in the brain. Deep neural networks consist of a series of stacked computations, sometimes including a nonlinear computation known as normalization. Normalization is a prevalent computation in models of the brain, and can be associated with a host of neural and perceptual effects, such as visual illusions. In this talk, I will describe how we are building richer normalization models into deep neural networks, inspired by our understanding of primary visual cortex in the brain. I will discuss how this work can have mutual impacts to inform models of the brain and to advance artificial intelligence.