ILPnet2 is the Inductive Logic Programming (ILP) Network of Excellence consisting of over 20 universities and research institutes. In addition the network actively pursues industrial relations through its End-User Club (EUC), consisting of companies and other non-academic institutions interested in practical applications of ILP. ILPnet2 succeeds ILPnet, the ILP European Scientific Network.
The mission of ILPnet2 is defined by the following long-term objectives:
The workplan for the third year was as follows:
Below we report on the activities of ILPnet2 in the third year and relate them to the workplan. These include the publication of the ILP Newsletter, significant updates of the WWW resources (library, systems and application descriptions), the organization of two data analysis workshops for members of the EUC, supporting the organization of three workshops (LLL-2000, ILP-2001, LLL-2001) and the organization of two educational events. Progress has also been made towards the establishment of an ILP toolbox for educational purposes. More detailed information on ILPnet2 activities can be found starting at the ILPnet2 website at http://www.cs.bris.ac.uk/~ILPnet2/.
Two data analysis workshops for end-users took place in year three. The first workshop took place in Leuven, Belgium, in October 2000. Hendrik Blockeel (KUL) and Saso Dzeroski (IJS) analyzed data on the water quality of Slovenian rivers in cooperation with the river ecologist Jasna Grbovic (EUC member - Hydrometeorological Institute of Slovenia). Logical decision trees were applied to predict river water biological communities from physical and chemical data. Jasna Grbovic presented the results of this analysis at the Second Symposium for European Freshwater Sciences.
The second data analysis workshop was held in Ljubljana in April 2001. Saso Dzeroski (IJS) worked together with Zdenek Zabokrtsky from the Center for Computational Linguistics (joined the EUC in year three) on several problems related to tectogrammatical annotation of Czech linguistic resources (automated functor assignment, resolving functor ambiguity for prepositional groups, and determining valency frames for verbs). Three propositional learning systems and one ILP system were applied to the data, which encompassed tectogrammatical trees from the Prague Dependency Treebank project, the EuroWordNet lexical database, and a lexicon of verb valency frames.
In the third year, ILPnet2 supported two ILP-related educational events.
The complete materials (slides, accompanying papers) from the three abovementioned lectures are available in the common electronic workspace; the slides are also available on the Web.
Three workshops were supported: LLL-2000 and LLL-2001, as well as ILP-2001.
An ILP educational toolbox is being developed by the University of Magdeburg (taken over from GMD). The toolbox is based on the data mining toolbox KEPLER, developed by GMD and DIALOGIS. It will include tutorial cases of using PROGOL and FOIL on illustrative problems as well as a realistic problem, i.e., the mutagenesis problem.
Due to problems in the operation of DIALOGIS, which closed down in May 2001, we have experienced some delays in the development and release of the ILP educational toolbox. However, the minor changes of KEPLER necessary for the toolbox have now been made and the tutorial materials (data, guided tour) for PROGOL in KEPLER have already been prepared. The resulting release of KEPLER is now being tested and will be distributed to ILPnet2 nodes before the end of 2001. The data and tutorial materials for FOIL in KEPLER will be prepared and distributed in early 2002.
ILPnet2 exploits the Internet as the main medium of information dissemination. Information disseminated in this way includes bibliographic information (also technical reports and thesis abstracts), information on ILP systems and information on ILP applications and datasets. In this respect, ILPnet2 expands on activities undertaken by ILPnet, the ILPnet2 predecessor.
In year three, many new entries were submitted to the ILPnet2 on-line library. These include, among others, the papers from the workshops sponsored by ILPnet2 (LLL-2000 and LLL-2001, ILP-2001), articles from two special issues of the Machine Learning Journal on ILP, and chapters from several ILP-related books. The library now contains over 1000 entries.
Information on ILP systems was substantially expanded in year three: 12 new entries were added, and 6 entries updated. There are now almost 50 entries in the ILPnet2 systems page. Progress was also made on testing the portability of ILP systems: information on the portability of 8 ILP systems is now available.
The ILPnet2 applications and datasets page was also updated: 9 new entries were added and 3 existing entries updated. Individual application descriptions were regrouped again to facilitate inspection by potential end-users. Most of the new entries concerned applications of ILP in natural language processing.
Other information available through the ILPnet2 website includes the ILP newsletter, a list of ILP-related events and a list of educational materials. The ILP Newsletter had three issues in the third year of ILPnet2. In addition to being distributed via email, the newsletter is also available through the ILPnet2 website at IJS. The list of ILP-related events, created in year two, was maintained and contains over 20 events for 2001. The list of educational materials, which include on-line versions of lectures/slides of ILP related courses or tutorials, was updated with several new entries.
The ILPnet2 website and its contents are described in detail in a chapter of the book Relational Data Mining, edited by Saso Dzeroski and Nada Lavrac (published by Springer in September 2001). The book gives a comprehensive overview of ILP methods relevant for knowledge discovery in databases, as well as applications thereof. The chapter, written by Todorovski et al., will contribute to the promotion of ILPnet2 and ILP.
In summary, all of the activities planned in the workplan for the third year have taken place in their originally planned or a similar form. Two data analysis workshops for end users took place and there was substantial progress in maintaining and extending the various kinds of information through the WWW. There is some delay with the deployment of the ILP educational toolbox, but the toolbox is almost ready and the one year no-cost extension will enable us to disseminate the toolbox and carry out this activity to its full extent. We did not have a separate summer school on ILP and its applications, but had two other educational events. In particular, a tutorial on rule induction at IDA-2001 and a session on machine learning and ILP in the context of agents at ACAI-2001.
In the final year, we are planning to extend ILPnet2 with several new nodes/associate nodes. We are also planning to make some additions to the End-User-Club and conduct at least one data analysis workhop for end users. Finally, we are planning a summer school on ILP and its applications (e.g., Relational Data Mining).