Céline Rouveirol and Michèle Sebag
The eleventh international conference on Inductive Logic Programming, ILP-01, was held at Strasbourg, France, on September 9-11th, 2001. ILP-01 was collocated with the 3rd international workshop on Logic, Learning and Language (LLL-01), and nearly collocated with the joint 12th European conference on Machine Learning (ECML-01) and 5th European conference on Principles and Practice of Knowledge Discovery in Databases (PKDD-01).
Continuing a series of international conferences devoted to Inductive Logic Programming and Relational Learning, ILP-01 was the central event in 2001 for researchers interested in learning structured knowledge from structured examples and background knowledge.
Lately, one of the major challenges for ILP has became to contribute to the exponential emergence of Data Mining, and to address the handling of multi-relational databases. On one hand, ILP has developed a body of theoretical results and algorithmic strategies for exploring relational data, essentially but not exclusively from a supervised learning viewpoint. These results are directly relevant to an efficient exploration of multi-relational databases and/or use of background knowledge.
On the other hand, Data Mining might require specific relational strategies to be developed, especially with regard to the scalability issue. The near-collocation of ILP-01 with ECML-01/PKDD-01 was an incentive to cross-fertilization between the ILP relational savoir-faire and the new problems and learning goals addressed and to be addressed in Data Mining.
The invited talks, respectively given by Dan Roth, Univ of Illinois, USA, and
H. T.T. Toivonen, Nokia, Finland, described the challenges
in two among the hottest fields for Machine Learning
and ILP: Natural Language, and the Genome (presentations will be made
available on ILP-01 Web site
Thirty seven papers were submitted to ILP, among which twenty one have been
selected as regular papers. They appear in the proceedings, published
by Springer Verlag LNCS LNAI 2157, URL
Several non-disjoint trends are observed, along an admittedly subjective clustering.
On the theoretical side, a new inference framework unifying induction, abduction and consequence finding, is proposed by Katsumi Inoue. New learning refinement operators are proposed by Liviu Badea, while Ramon Otero investigates negation-handling settings.
Several hybrid frameworks are proposed, either bridging the gap between ILP and other learning paradigms, e.g. Bayesian inference (Kristian Kersting and Luc De Raedt), Neural Nets (Rodrigo Basilio, Gerson Zaverucha, and Valmir C. Barbosa) or Feature Selection (Tomosobu Ozabaki and Koichi Furukawa) or exploiting other search paradigms, e.g. Constraint Satisfaction (Jérome Maloberti) or Genetic Algorithms (Agnès Braud and Christel Vrain), to address particular ILP tasks.
Among the tasks addressed, changes of representation take an increasing importance, ranging from propositionalization (Agnès Braud and Christel Vrain, already mentioned) and construction of structural features (Stefan Kramer), to aggregation-based transformations (Mark-A. Krogel and Stefan Wrobel).
Committee-based and statistical machine learning interestingly pervade ILP through boosting-based approaches (Susanne Hoche and Stefan Wrobel), positive-only learning (Filip Zelezny), and transductive inference (Martin Eineborg and Henrik Boström, already mentioned). Last but not least, an efficient relational cross-validation procedure is proposed by Jan Struyf and Hendrik Blockeel.
The application papers deserve a special mention as they demonstrate
when and how relational representations can make a difference.
Language-related applications range from Natural Language (Miloslav Nepil)
to XML documents (Akihiro Yamamoto, Kimihito Ito, Akira Ishino,
Hiroki Arimura), and shell logs (Nico Jacobs and Hendrik Blockeel).
Bio informatics offers many challenging relational problems
(Andreas Karwath and Ross D. King), in the spirit of the founding
ILP application, i.e. the mutagenesis problem.
Other applications are concerned with medical control
(René Quiniou, Marie-Odile Cordier, Guy Carrault, Feng Wang)
and spatial data mining (Donato Malerba and Francesca A. Lisi).
In addition to the presentation of regular papers, ILP-01 also had a work-in-progress track showcasing promising new directions and allowing feedback on preliminary work. Eleven posters were presented in this track with associated papers published in a Technical Report of the Strasbourg University. This volume will also be made available on the world wide web.
ILP-01 would not have been possible without the generous support of several
sponsors, whom we could like to thank on behalf of the whole ILP community:
ILPNet2: The European Network of Excellence in Inductive Logic Programming
Université Robert Schuman
Université Louis Pasteur
Conseil général du Bas-Rhin
IUT Robert Schuman,
and the Strasbourg City.
Further information on ILP2001 can be found at
ILP2002 will be held in Sydney, Australia, during the summer of 2002.