Classifier Chains for Multi-label Classification

Title: Classifier Chains for Multi-label Classification

 

Abstract: Work on multi-label classification in the academic literature has expanded prolifically in recent years. In the multi-label scenario, each data instance may be associated with multiple class labels simultaneously (as opposed to a single class label, as in the traditional classification task). This context arises naturally and frequently in many domains: text categorisation, tagging images and other media, medical diagnosis, and bioinformatics, to name a few. The main challenge is to model the dependencies among labels, which must be done efficiently for the learning algorithm to scale up to large datasets. This talk will review advances in multi-label learning with 'classifier chains'; an approach which has attracted recent interest and yields high performance in a variety of multi-label classification tasks. Connections will also be made to related areas like structured output learning and probabilistic graphical models.

 

Bio: Jesse Read has recently joined Aalto University (in December 2013). Prior to that he was a lecturer, and earlier a postdoc, at the University Carlos III of Madrid, Spain. In 2010 he completed his PhD in the University of Waikato in New Zealand, with a thesis on Scalable Multi-label Classification. He has continued to work on multi-label classification and related tasks such as multi-output prediction; but has also published in the areas of data-stream classification, and wireless sensor networks.