Christoph Lampert
(http://pub.ist.ac.at/~chl/)
Institute of Science and Technology (IST) Austria
Tuesday 18th September 2012
Time: 4pm
4th Floor Seminar Room,
Alexandra House, 17 Queen Square, London, WC1N 3AR
Semantic Attributes and Classification without Training Examples
Can we use machine learning to build classifiers for categories for which we don't have any training examples? No, seems to be the obvious answer. However, in my talk I will discuss a setting that does indeed allow non-trivial learning in this situation, where the main idea is to rely on class descriptions in terms of semantic attributes. Using examples from recent work in computer vision I will explain the two main models in the field, direct attribute prediction and indirect attribute prediction, as well as recent extensions.