Prof. Tal Hassner (The Open University of Israel) is visiting Imagelab and is giving a talk on
" Faces, deep learning and the pursuit of training data"
Tuesday May 17, 2016 - 02:00 p.m. - Aula P 0.2 ex FA-0A
Abstract: The abilities of machines to detect and recognize faces improved remarkably over the last few years. This progress can at least partially be explained by the sizes of the training sets used to train deep learning models: huge numbers of face images downloaded and manually labeled. It is not clear, however, if the formidable task of collecting and labeling so many images is truly necessary. I will discuss the problems of data collection and describe a number of effective techniques for maximizing deep learning capabilities when collecting additional data is not an option. Importantly, though this talk will focus on face processing related tasks, these techniques can be applied in other image understanding problems where obtaining enough labeled examples for training deep learning systems is hard.