Universal, unsupervised, and understandable deep image representations
Monday, June 26th, 11am
Speaker: Andrea Vedaldi, associate professor at University of Oxford, U.K.
Abstract: Modern deep neural networks have taken computer vision by storm. In this talk, I will demonstrate a few applications of large scale neural nets to problems such as object recognition and text spotting. I will also discuss one of the main difficulties in using these models in new applications, namely the need of providing very large datasets of annotated images. I will show how this problem can be alleviated by the use of synthetic data. Then, I will discuss some of our recent research at the core of this technology, aimed at solving some of these problems at a more fundamental level. I will focus on three areas: unsupervised representations, universal representations, and understanding representations.