Tutorial by

Grigorios Tsoumakas (Aristotle University of Thessaloniki) , Min Ling Zhang (Hohai University) and Zhi-Hua Zhou (Nanjing University)

Monday 7  September (Room Kosovel, Hotel Park)

Abstract:

The tutorial on “Learning from Multi-label Data” is composed of four parts that cover the basic concepts as well as the state-of-the-art in multi-label learning. The first part gives a formal definition of multi-label learning, presents a rich variety of  applications and real-world datasets and describes commonly-used evaluation metrics. In the second part, existing multi-label learning techniques are reviewed in two groups. The first group of methods solve the task of multi-label learning by transforming it into one or more single-label classification tasks, while the second group of methods work by adapting existing single-label learning algorithms to deal with multi-label data. In the third part, two advanced topics on multi-label learning are discussed, i.e. learning in the presence of label structure and multi-instance multi-label learning (MIML). Finally, the fourth part presents the Mulan open-source library for multi-label learning.