Multilabel Rule Learning

Inductive rule learning is a very traditional, well-established approach in machine learning. Rule learning algorithms are typically employed when one is not only interested in accurate predictions but also requires an interpretable theory that can be understood, analyzed, and qualitatively evaluated by domain experts.

In this project we aim at solving multilabel classification (MLC) problems using rule learning algorithms. MLC has received a lot of attention in the recent machine learning literature and is nowadays used in applications as diverse as music categorization, semantic scene classification, or protein function classification. MLC becomes particularly challenging when it comes to discovering hidden dependencies between labels. As rules provide a natural form of expressing such dependencies, they are well-suited to discover label correlations in a human-comprehensible manner.

The objectives of this project are to develop a unified framework for representing different types of label dependencies using rules and to analyze their expressive power for MLC problems. Furthermore, we will investigate approaches to overcome the algorithmic challenges of learning such multilabel rule sets from data and we will evaluate the predictive and descriptive performance of such models in comparison to state-of-the-art systems.

Publications