C o m p u t a t i o n a l    L o g i c

The 7th International Workshop on Inductive Logic Programming (ILP-97)


The 7th International Workshop on Inductive Logic Programming (ILP-97)

Simon Anthony

Few visitors to Prague could fail to appreciate its beautiful architecture and rich cultural history. Between the 15th and the 20th of September, these inspiring surroundings provided the location for the following Inductive Logic Programming (ILP) events:

* the 1st ILP Summer School, entitled "Inductive Logic Programming and Knowledge Discovery in Databases'',

* the 7th International Workshop on Inductive Logic Programming (ILP-97), and

* the 1st meeting of the CompulogNet Area "Computational Logic and Machine Learning'', entitled "Representation Issues in Reasoning and Learning''.

This week of events was locally organised by Olga Stepankova and her colleagues at the Czech Technical University, and allowed 77 researchers to participate in a stimulating and varied programme. The week was financially supported by the kind sponsorship of two European Networks of Excellence: MLNet and CompulogNet, the End User Club of the ILP2 projects, as well as the Czech Technical University and the J. Stefan Institute.

ILP-97 Workshop

The ILP-97 programme, chaired by Nada Lavraťc and Saso Dzeroski, described theoretical, empirical and applied research in ILP. 26 refereed papers were presented at 10 themed sessions, together with 3 additional talks given by invited speakers. Brief descriptions of these invited talks are given below.

Usama Fayyad, of Microsoft Research, gave the first invited talk on the emerging discipline of Knowledge Discovery in Databases (KDD). The large amounts of data now being stored in relational databases has prompted researchers to explore the potential of "mining'' this data in order to discover hidden patterns and relationships. For instance, the use of KDD methods has allowed astronomers to improve their accuracy in distinguishing stars from galaxies, as well as discovering several new quasars. It is within this "Data Mining'' step of the KDD process that ILP has a potentially key role to play, particularly due to its powerful yet perspicuous first order logical representation. However, in order to achieve this potential, ILP must face a number of challenges such as the need to handle very large datasets.

In the second of the invited talks, Jean-Francois Puget of ILOG discussed the possible benefits of combining the key ideas of Constraint Logic Programming (CLP) and ILP. CLP provides the ability to constrain variables (instead of fully instantiating them) and uses a logical handling of numbers, which eases some of the number handling problems faced by conventional logic programming. ILP may be able to contribute to CLP by:

* learning the cause of query failure by identifying small sets of 'culprit' variables, and

* generalised constraint propagation by finding a simple 'approximation' constraint that is a least upper bound for a set of computed answers to a query.

Conversely, CLP can return these favours by:

* providing a CLP representation language and therefore a logical handling of numbers, and

* viewing an ILP learning task as a Constraint Satisfaction Problem and using existing CLP techniques in order to find a solution.

The challenge to the ILP community is to employ CLP as a representation language and, in return, to examine the contributions ILP can make during query execution by CLP systems.

The third and final invited talk, given by Georg Gottlob of the Vienna Technical University, addressed the complexity of some ILP problems. A useful first step when attempting to solve any computational problem is to study its complexity (or difficulty). Such a study can allow:

* a systematic approach to finding tractable subclasses of problems by identifying the sources of complexity, and

* the problem to be classified as belonging to a particular complexity class. Algorithms used to solve other problems in that class may shed light on how the problem in question might be solved.

From an ILP perspective, it has been shown that a logic programming formalism can be specified for any particular complexity class. However, many of the representations actually used in ILP are members of either the hard or very hard classes. The resulting conjecture is that the design of ILP algorithms should be increasingly based upon a complexity analysis of the learning task that they are to perform.

The technical programme consisted of 26 papers.

In the proceedings, published by Springer-Verlag,

the invited lectures and 9 selected papers are published as long papers, and the remaining 17 papers appear as short papers.

Simon Anthony

Department of Computer Science,

University of York,

York.

YO1 5DD, U.K.

Email: simona@cs.york.ac.uk


ILP98

The Eighth International Conference on Inductive Logic Programming (ILP'98) will be held July 22-24. ILP'98 will be co-located in Madison, Wisconsin, U.S.A., with the International Conference on Machine Learning (ICML'98) and five other related conferences (see list below). Original papers in all areas of ILP are solicited. The submission deadline is February 25, 1998.

Submission Requirements: Four hard copies or one electronic (printable postscript) copy of manuscript (10 pages maximum) received by the program chair on or before Feb. 25, 1998. Electronic submission is preferred. Please use subject "ILP98 Submission".

Program Chair: David Page

Speed Scientific School

University of Louisville

Louisville, KY 40292

U.S.A.

cdpage01@homer.louisville.edu

Co-located Conferences:

AAAI-98 (National Conference on Artificial Intelligence)

CogSci'98 (Cognitive Science)

COLT'98 (Computational Learning Theory)

GP-98 (Genetic Programming)

ICML'98 (International Conference on Machine Learning)

UAI'98 (Uncertainty in Artificial Intelligence)

Further information can be obtained from the following URL

WWW URL: http://www.cs.louisville.edu/faculty/page/ilp98


Coordinator's Report ] Computational Logic and Machine Learning ] BOOK ANNOUNCEMENT ] [ The 7th International Workshop on Inductive Logic Programming (ILP-97) ] Biomedical Applications of Computational Logic and Machine Learning ] Data Mining and Knowledge Discovery ] International Summer School on Inductive Logic Programming and Knowledge Discovery in Databases ] Frontiers of Inductive Logic Programming ] Abduction and Induction in AI ]


Home ] Automated Deduction Systems ] Computational Logic & Machine Learning ] Concurrent & Constraint Logic Programming ] Language Design, Semantics & Verification Methods ] Logic Based Databases ] Program Development ] Knowledge Representation & Reasoning ]