Rule learning has a long history within the field of machine learning. In particular, the so-called separate-and-conquer or covering family of rule-based classification algorithms goes back to the early days of machine learning. Since the first papers on the AQ rule learning algorithm, this research area has had its ups and downs but it never completely vanished from the research menus of our field. The primary reason for this lies in the attractiveness of rules as the arguably most comprehensive concept representation formalism. After its peak in the early nineties (through the advent of Inductive Logic Programming algorithms), the focus of research in rule learning soon shifted to association rule discovery, and interest in inductive rule learning declined. Lately, we can observe another increase of interest in this area. Recent advances include novel methods for handling contradicting or missing predictions, multi-instance rule learning, subgroup discovery, integration of boosting and covering, covering on association rules, statistical approaches to rule-based prediction and clustering, efficient learning with rule templates, alternatives to the covering strategy improved strategies for handling multi-class problems, meta-learning of rule evaluation metrics, and many more (references to papers on these topics can be found in the call for papers).
The presentations at this workshop represent a snapshot of ongoing work in this area, and touch upon many of the above-mentioned topics. The breadth of problems addressed in these works underlines that despite its long history, rule learning is far from being a well-understood problem. We believe the time was right for a workshop devoted to this ancient topic in machine learning.
The papers presented at this workshop offer a representative snapshot of ongoing work in this area, and touch upon many of the above-mentioned topics. The breadth of problems addressed in these works underlines that despite its long history, rule learning is far from being a well-understood problem. We believe the time was right for a workshop devoted to this ancient topic in machine learning.