LG204: Feature Learning - Theory, Practice and Applications

Date: 
Friday, September 26, 2014 - 12:00
Speaker: 
Angel Cruz Roa, PhD & Haibo Wange, PhD
Abstract: 
Feature representation is the most fundamental issue in computer vision. Recently, hierarchical feature learning, a.k.a. deep learning (DL), emerges as the primary solution for this issue due to its successful application in several tasks such as automatic image annotation and character recognition by Google Brain and other IT companies like Facebook, Microsoft, and Baidu. In this talk we will briefly introduce deep learning from both a theoretical and practical perspective. In the first part of this talk, Haibo Wang will present what is deep learning, why it's so effective, and which deep learning frameworks exist. In the second part, Angel Cruz will present Convolutional Neural Networks (CNN), the most widely used DL framework in computer vision. Meanwhile, Angel will present how to use Torch7, one of the best deep learning frameworks that is publicly available. Finally, we will present two example applications of DL in breast cancer image analysis.