VR-43-XN. ISSN Volume, 17, number 3 June, Wiebe van der Hoek (member) Radboud Winkels (section editor)

ISSN 1566 - 8266

Volume, 17, number 3 June, 2000

BOARD BNVKI: Joost Kok (chair) Rineke Verbrugge (member) Wiebe van der Hoek (member) Yao-Hua Tan (member) Eric Postma (member) Luc DeHaspe (member) Walter Daelemans (member) Gert-Jan Beijer (member)

BNVKI-SECRETARY: Rineke Verbrugge Rijksuniversiteit Groningen Cognitive Science and Engineering Grote Kruisstraat 2/1 9712 TS Groningen [email protected]

EDITORIAL BOARD: Eric Postma (Editor in Chief) Jaap van den Herik Bart de Boer Shan-Hwei Nienhuys-Cheng Cees Witteveen Antal van den Bosch (section editor) Edwin de Jong (editor Belgium) Richard Starmans (section editor) Radboud Winkels (section editor) EDITORIAL ADDRESS BNVKI newsletter Joke Hellemons Universiteit Maastricht FdAW, Department of Computer Science P.O.Box 616, 6200 MD Maastricht Telephone: 043 388 34 77 / 35 04 Fax: 043 388 48 97 E-mail: [email protected] http://www.cs. unimaas.nl/~bnvki

Segmentation Unit Character Isolator Character Recognizer

VR-43-XN

Syntactical Analyzer

KEEP IT SIMPLE, STUPID EDITOR-IN-CHIEF

Trivial ideas can have an enormous impact when applied in a new context. Consider the case of a Dutch teacher who had a trivial idea in the context of the Internet. He created a webpage called the Start Pagina (Starting Page) containing a list of hyperlinks to often-wanted sites, such as the telephone book, travel information agencies and so forth. Despite its simplicity, the Start Pagina is one of the most popular sites in the Netherlands. Both newcomers and experienced users of the web consult the Start Pagina for finding information amidst the chaos of the Internet. The interested reader may take a look at http://www.startpagina.nl. Statistics on the popularity of the site (and other sites) can be found at the NetStat page http://nl.nedstatbasic.net/. Recently, a large Dutch publishing organisation bought the Start Pagina for several tens of millions of guilders. The interesting aspect of the success of the Start Pagina is that it is based on an extremely simple idea. The power of simplicity is often emphasised in the context of (scientific) writing. KISS is a well-known acronym that refers to the phrase “Keep It Simple, Stupid" and indicates that simple formulations are to be preferred over difficult ones. The KISS principle applies to science and technology as well. Whereas, in modelling and theorising Occham’s Razor should be applied to keep the complexity of the model or theory as simple as possible, in the development of innovative ideas the KISS principle serves as an apt guideline. AI researchers think very hard about the problem of dealing with the information overload on the Internet. Their answers are sophisticated and potentially powerful. However, as of yet their impact on daily Internet usage is very small when compared to the impact of the Start Pagina or similar sites. Applying the KISS principle to Internet research facilitates a matching of technology to the needs of Internet users. Trivial ideas are not always as trivial to implement as the Start Pagina. Therefore, they may still pose technological challenges for applied AI research. The twelfth Belgium-Netherlands Conference on Artificial Intelligence (BNAIC 2000) will be held on November 1 and 2 (see the Call for Papers and Demonstrations on page 32 of this newsletter). As last year, there is a special submission class for applications and demonstrations. Simple and trivial ideas that are turned into successful AI applications can be submitted in this class. In case your idea is not in that stage yet, you may consider to write a paper about it and submit it to the regular-papers or compressed-contributions class. The deadline for submissions is May 31, 2000. For more information on the BNAIC 2000 or other AI events, your “Start Pagina” is http://www.cs.unimaas.nl/~bnvki.”

Cover photograph: Longneck (Langnek) in the Efteling near Tilburg, the location of the BNAIC 2000. See the Call for Papers on page 32 of this newsletter.

BNVKI newsletter

30

April 2000

TABLE OF CONTENTS Keep it Simple, Stupid (Editor-in-Chief) ...............................................................................................................30 Table of Contents ...................................................................................................................................................31 BNVKI-Board News (Joost Kok) .........................................................................................................................32 Call for Papers BNAIC 2000 (Antal van den Bosch) ...........................................................................................32 Reports BNAIC’99 (Gertjan Beijer) ......................................................................................................................33 A View on Ph.D.s (Jaap van den Herik) ................................................................................................................35 Thesis Reports........................................................................................................................................................35 Classification using Decisions Trees and Neural Nets (Wim Pijls) ................................................................35 Local Learning Techniques for Modelling, Prediction and Control (Yang Quan Chen) ................................37 SIKS (Richard Starmans).......................................................................................................................................39 Section Knowledge Systems in Law and Computer Science (Radboud Winkels).................................................39 Verslag Wetenschappelijk deel Jurix Bijeenkomst 9 februari 2000, UvA (Luuk Mathijssen)........................39 Jurix and Public Counter 2000 (Ronald Leenes)............................................................................................41 Knowledge Management (in een Juridische Context) (Martin Apistola).......................................................43 The Evolution of Language Conference (Bart de Boer) ........................................................................................45 Section AI in Business ...........................................................................................................................................46 Towards an Online Distribution Structure (Kees Jonkheer) ...........................................................................46 Call for Participation..............................................................................................................................................49 Playstation meets organism: emulation in a cultural context .........................................................................49 Conferences, symposia, Workshops.......................................................................................................................50 Emailaddresses, Board Members/ Editors BNVKI newsletter / How to become a member?/ Submissions..........51 Advertisement ........................................................................................................................................................52

The BNVKI is sponsored by AHOLD and by BOLESIAN In 2000, the publication of the BNVKI newsletter is also supported by the Division of Computer Science Research in the Netherlands (previously called SION, now ACI)

BNVKI newsletter

31

April 2000

BNVKI BOARD NEWS

CALL FOR PAPERS AND DEMONSTRATIONS

Joost Kok Chairman BNVKI

Antal van den Bosch Last week I visited the ACM Symposium on Applied Computing in Como, Italy. Although the weather and the food were nice, the town Como is a bit boring at the end of the winter. The conference was held in a beautiful villa at lake Como. Last year, a G7-meeting was held in the same villa. The conference was interesting and contained a large track on evolutionary algorithms.

TWELFTH BELGIUM-NETHERLANDS CONFERENCE ON ARTIFICIAL INTELLIGENCE

November 1-2, 2000

De Efteling, Kaatsheuvel http://ilk.kub.nl/bnaic00/

In addition, there were discussions on Computing Curricula 2001, organised by The Joint Task Force on Computing Curricula of the IEEE Computer Society and the Association for Computing Machinery. The last curriculum was proposed in 1991 and every ten years a new curriculum is put together. Citing the Computing Curricula website:

The twelfth Belgium-Netherlands Conference on Artificial Intelligence (BNAIC’00) is organised by the Infolab and the Induction of Linguistic Knowledge (ILK) research group at Tilburg University. The School for Information and Knowledge Systems (SIKS) has given its auspices to the conference. BNAIC is the annual conference of the BNVKI/AIABN, the Belgian-Dutch Association for Artificial Intelligence, and will be held at the Efteling, Kaatsheuvel, on November the 1st and 2nd, 2000.

“The Joint IEEE Computer Society/ACM Task Force on the Year 2001 Model Curricula for Computing (CC-2001) was formed to review the 1991 curricula and develop a revised and enhanced version for the Year 2001 that addresses developments in computing technologies in the past decade and will sustain through the next decade.”

The conference aims at presenting an overview of state-of-the artresearch in artificial intelligence in Belgium and The Netherlands, and to further the interaction between researchers in both countries. We have distinguished three classes of submission, each with its own refereeing procedure.

The draft version of the Curriculum proposal contains a section on Intelligent Systems, including topics such as Fundamental issues in intelligent systems, Search and optimisation methods, Knowledge representation and reasoning, Learning, Agents, Computer vision, Natural language processing, Pattern recognition, Advanced machine learning, Robotics, Knowledge-based systems, Neural networks, and Genetic algorithms. The draft report can be found at http://www.computer.org/education/cc2001/report/ indexe.html. It is still possible to give comments on the proposal. Contact addresses can be found on the web site.

The classes are A for regular paper submissions, B for compressed contributions, and C for demonstrations and applications. Class A. The essence of every conference is the presentation of new and original work. This class accommodates the submission of such work. Submitted papers should not exceed a length of 8 pages.

Finally, I would like to encourage our members to submit contributions reporting or demonstrating on their research progress to the BNAIC 2000, to be held at the Efteling in Tilburg. A beautiful location to listen to research, talk about research, and to dream of fairytales.

BNVKI newsletter

Class B. Work that has been reviewed and subsequently published elsewhere can be submitted as a compressed contribution. Authors are invited to submit the officially published version (without 32

April 2000

restrictions on the number of pages) together with a one or two-page abstract (to appear in the BNAIC proceedings).

PROGRAMME COMMITTEE

Antal van den Bosch (co-chair, KUB) Hans Weigand (co-chair, KUB)

Class C. Proposals for demonstrations will be evaluated based on submitted demonstration summaries (in English) stating the following: the purpose of the system to be demonstrated, its user groups, the organisation or project for which it is developed, the developers, and the technology used. In addition, the system requirements and the duration (not exceeding 30 minutes) should be mentioned. Especially researchers from industry are encouraged to submit their successful applications to this class. The maximum size of demonstration summaries is 2 pages.

CONTACT INFORMATION

Alice Kloosterhuis (KUB) Infolab, Tilburg University P.O. Box 90153 5000 LE Tilburg The Netherlands Email: [email protected] Telephone: +31 13 466 3020 Fax: +31 13 466 3069

For all classes, possible topics of submissions include, but are not limited to: multi-agent systems, neural networks, knowledge-based systems, natural language processing, games, search, machine learning, robotics, knowledge representation, knowledge management, ontologies, logic programming, optimisation, intelligent agents, and evolutionary algorithms. CLIN

IMPORTANT DATES

Deadline for submissions: May 31, 2000 Notification of acceptance: July 15, 2000 BNAIC’00 HOMEPAGE For all up-to-date information on BNAIC, please visit (and bookmark) http://ilk.kub.nl/bnaic00/

2000

Immediately after BNAIC’00, on November 3, 2000, the eleventh Computational Linguistics in the Netherlands meeting (CLIN) will be held in Tilburg. CLIN is organised by a team headed by Walter Daelemans (UIA Antwerpen, KUB). The overlap in research interests of the BNVKI and CLIN communities will be strengthened furthermore by an invited speaker (Yorick Wilks) and a special BNAIC session on computational linguistics / language technology.

REPORTS BNAIC’99 DEMONSTRATIONS

Report by Gertjan Beijer Bolesian

I hereby present my postponed summary of the session “Demonstrations 1” presented during the BNAIC 1999 (!), which was held in Maastricht last year. All three contributions consisted of a short presentation on the backgrounds followed by a demonstration.

SUBMISSIONS

Papers and demonstration summaries are to be submitted as postscript file and sent by email to Alice Kloosterhuis ([email protected], +31 13 466 3020). A LaTeX style and a MS Word template will be made available through the BNAIC’00 Web site (see below). All submissions should be accompanied by a message stating the submission class (A, B, or C). Proper receipt of submissions will be acknowledged by e-mail. The deadline for submissions is May 31, 2000. Accepted papers will be published in the Proceedings. Authors keep the copyright of their submissions.

IGUANA: A WEB CRAWLER AIMED AT CREATING A DOMAIN SPECIFIC WEB ENGINE

R.. Ekkelenkamp TNO-TPD

Mr. Van Breukelen presented this session’s lecture and demo instead of the author is R. Ekkelenkamp.

INVITED SPEAKERS

Yoav Shoham (Stanford University) Yorick Wilks (University of Sheffield)

BNVKI newsletter

33

April 2000

Iguana is a webcrawler aimed at crawling domain specific websites. In order to make this possible the program contains a number of text categorisation algorithms to perform its main task. Amongst them are vector-space index search and fuzzy noun phrase index-based search.

Neural Vision is a software package intended for exploratory data analysis using projection techniques to display multidimensional data on two-dimensional screens. The techniques used come from neural network theory and statistics. With Neural vision one can visually discover clusters of data or relationships between data-elements before deciding to use advanced (i.e., costly and time-consuming) analysis techniques to (hopefully) discover new information or knowledge.

Iguana, which was written in Java, has proven its value in a number of applications, amongst others for the Dutch Milleniumplatform and a community of a number of European cities. That is why its core engine contains features for automatic language understanding and disposes a user interface in four languages.

The sheets presented some backgrounds on the application such as the partners who worked on the project (RUL, CWI, RWS) and a short history (it all started in 1996…). Techniques applied are Sanger’s Neural-Network based Principal Components Analysis (PCA) projection technique yielding the maximal amount of variance in a small number of dimensions, Independent Components Analysis (ICA) as a second projection technique that goes further than Principal Components Analysis (it tries to find dimensions that are statistically independent) and finally a Frequency Sensitive Competitive Neural Network which divides input space in clusters minimising within-cluster variance.

The actual demo showed a number of Iguana’s features like searching on phrases, finding (translated) German webpages, sorting on suitability and the “search similar function”. Text mining is a hot topic in academic and industrial environments. It is still hard to prove that a certain approach is the best regarding the characteristics of the data, such as distribution of items, distribution of categories, subject of text etc etc. It is still hard to even simply compare all different approaches! At the end of the demonstration questions were asked about the measures for calculating similarity and distance between documents. Unfortunately there was not much time left.

The application was written in Borland C++ Builder 3.0. The demo showed the application together with two data sets. One data set was used for the classification of Iris flowers. Training and testing was demonstrated for this 4 dimensional data set. The second data set represented various types of data regarding roads, amongst others its use and maintenance costs. This 11 dimensional data set was used for a number of visual explorations with colourings, polygonplots and additional statistics.

NEURAL VISION 2.0

Michiel van Weezel CWI

The winner of the Best Demonstration Award was Michiel van Weezel with his lecture about Neural Vision 2.0. This demonstration was chosen to be the best because it most clearly demonstrated how advances in science can almost be put into “everyday” practice. BNVKI newsletter

AGENT BASED CUSTOMER SERVICE

34

April 2000

agents applications are developed in Java and Jess (Java expert system shell). A new development area entering KPN research is machine learning in combination with agent technology.

Jan Heemskerk KPN research

The third and final session of the “Demonstrations 1” sessions was chosen second best in the Best Demonstration Competition for it’s clarity and was presented by Jan Heemskerk of KPN Research.

A VIEW ON Ph.D.s Jaap van den Herik IKAT, Universiteit Maastricht Within the period 1995 – 2005, the year 2000 will not be the most productive year of all Ph.D. students. It seems that the millenium transition has encouraged many people to finish their Ph.D. research still in the previous century. So be it. A new generation starts slowly and I assume that we have to wait until 2002 before we are back on the old level of approximately 25 Ph.D. students a year.

His short and professional presentation contained some background of the company he worked for which provides work for 500 researchers! Jan is primarily focussed on agent-based technology. Within KPN he is responsible for the agent-based research program which consists of 12 projects. The demonstration describes one of these projects. The business environment in which the solutions should work is today’s customer contact infrastructure. It depends on a technical system environment in which agent building blocks will finally get their place.

This time we are happy to announce four defences. They are all four from different areas, namely logic, software engineering, modelling, and knowledge retrieval. Niesink and Van der Pol are SIKSpromovendi. The Ph.D. theses of Kamps and Hulstijn are results of large projects which are carried out at the universities mentioned. The Editor congratulates three successful doctores and wishes the last one (Van der Pol) much strength with his defence. Moreover, we have two Ph.D. reviews. The first review is written by Wim Pijls and gives an account of the Ph.D. thesis by Rob Potharst (EUR) titled: “Classification using Decision Trees and Neural Nets”. The second review is written by Yang Quan Chen (and introduced by Edwin de Jong) and discusses Gianluca Bontempi’s thesis “Local Learning Techniques for Modelling, Prediction and Control.” We look forward to other Ph.D. reviews for the next issue. New announcements on Ph.D. defences are also welcome.

The demonstration showed some nice features of the highly-distributed solution by showing applications that are investigated and developed for interpreting customer questions, personalising information supply and serving as a onestop shop for the customer. Agent characteristics like autonomy, selfknowledge and knowledge about its environment emerge and are applied…

J. Kamps (March 10, 2000) A Logical Approach to Computational Theory Building. Universiteit van Amsterdam. Promotor: Prof.dr. W.E. Saris, copromotor: dr. J.M.F. Masuch.

A generic architecture, based on the FIPAarchitecture, is used in which messages from the user propagate through the system via an interface agent reaching different agents in the background. Natural language is translated to ACL and an ontology is used to disambiguate phrases. Each agent has its own problem solving task such as translating natural language to meaningful phrases, planning activities or determining the right answer to a posed question. The BNVKI newsletter

F. Niesink (March 28, 2000) Perspectives on Improving Software Maintenance. Vrije Universiteit Amsterdam. Promotor: Prof.dr. J.C. van Vliet. J. Hulstijn (April 7, 2000) Dialogue Models for Inquiry and Transaction. Universtiteit Twente. Promotor: Prof. dr.ir. A. Nijholt.

35

April 2000

R.W. van der Pol (September 14, 2000) Knowledge-based Query Formulation in Information Retrieval. Universiteit Maastricht. Promotores: Prof.dr. H.J. van den Herik, Prof.dr.ir. A. Hasman, and Prof.dr.ir. J.L.G. Dietz CLASSIFICATION USING DECISION TREES AND NEURAL NETS

classificeren. Dit onderwerp behoort tot het gebied van "supervised learning". Ofschoon beslisbomen reeds veel aandacht gekregen hebben in de literatuur, weet de auteur in genoemde hoofdstukken enkele belangwekkende nieuwe resultaten te presenteren. De hoofdstukken 5 en 6 hebben een andere invalshoek. In hoofdstuk 5 worden Bayesiaanse neurale netwerken besproken. In hoofdstuk 6 wordt een verband gelegd tussen beslisbomen en neurale netwerken.

Proefschrift van R. Potharst Erasmus Universiteit Rotterdam

Bespreking door W. Pijls In juni 1999 promoveerde Rob Potharst aan de Erasmus Universiteit Rotterdam op een proefschrift getiteld: "Classification using Decision Trees and Neural Nets". Promotor was Prof. A. de Bruin met als co-promotor en belangrijkste begeleider Dr. J.C. Bioch.

MONOTONIE

Na een inleidend eerste hoofdstuk, begint in het omvangrijke tweede hoofdstuk de bespreking van beslisbomen. Een belangrijk begrip in dit hoofdstuk is het begrip monotone dataset. Stel dat alle attributen (inclusief de klasse) ordinale waarden hebben. Een dataset heet monotoon, indien objecten met hogere attribuutwaarden ook tot een hogere (althans niet lagere) klasse behoren. Een classificatiesysteem heet monotoon, indien objecten met hogere attribuutwaarden ook hoger geclassificeerd worden. De meest bekende algoritmen voor het induceren van beslisbomen zijn ID3, C4.5 en CART. Deze hebben alle het bezwaar dat ze geen monotone beslisboom induceren, d.w.z. de beslisboom geïnduceerd door een monotone trainingsset garandeert geen monotone classificatie voor uitbreidingen van de trainingsset (bij voorbeeld een uitbreiding met de testset). In hoofdstuk 2 worden enkele nieuwe algoritmen gepresenteerd die wél monotone beslisbomen induceren. Dit alles wordt op een gedegen theoretische manier onderbouwd. Het nadeel van de nieuwe algoritmen is wel dat de beslisbomen groter zijn en dat het induceren van de bomen relatief duur is. Verder wordt in dit hoofdstuk ook nog een algoritme beschreven, dat nagaat of een gegeven beslisboom wel of niet monotoon

Zoals de titel reeds aangeeft, is classificatie het centrale thema in dit proefschrift. Onder classificatie wordt verstaan: het onderbrengen van objecten, op basis van de attributen van dit object, in één van een aantal vooraf gedefinieerde klassen. Twee bekende classificatietechnieken, te weten beslisbomen en neurale netwerken, worden bestudeerd, waarbij de eerstgenoemde techniek duidelijk meer aandacht krijgt dan de tweede. De classificatiesystemen zijn zogenaamde lerende systemen, d.w.z. ze classificeren op basis van indelingen in het verleden. De systemen krijgen een zogenaamde trainingsset aangeboden, een set van objecten gedefinieerd in termen van een aantal attributen. AFBEELDING VAN BESLISBOMEN OP NEURALE NETWERKEN

Beslisbomen spelen, zoals gezegd, een belangrijke rol in dit proefschrift. De hoofdstukken 2, 3 en 4 handelen over het construeren (ook wel aangeduid als het induceren ) van beslisbomen op basis van een gegeven dataset (de trainingsset). De beslisboom wordt gebruikt om de elementen van een andere set (de testset) te BNVKI newsletter

36

April 2000

classificeert. Alle algoritmen zijn uitvoerig experimenteel getest.

BAYESIAANSE NEURALE NETWERKEN

De laatste twee hoofdstukken richten zich op neurale netwerken als instrument voor classificatie. In hoofdstuk 5 wordt een zogeheten Bayesiaanse benadering voor het bouwen van classificerende netwerken besproken. Deze benadering zou gunstig zijn voor het voorkomen van overfitting. In het kader van zijn onderzoek heeft de auteur een volledig netwerk gebaseerd op de Bayesiaanse aanpak geïmplementeerd. Op zich een taak waarmee je heel wat overhoop haalt. Als apriori-verdelingen worden standaard-normale verdelingen gekozen. Voor de berekening van de aposteriori verdelingen moeten enkele numerieke integralen opgelost worden. Hiervoor wordt de recent geïntroduceerde Markov Chain Monte Carlo (MCMC) methode gebruikt.

QUASIMONOTONIE

Ofschoon men monotone datasets niet zeldzaam kan noemen, is het construeren van een monotoon classificatiesysteem toch een zware eis. Daarom wordt het begrip quasimonotonie geïntroduceerd, een uitbreiding van het begrip quasipositiviteit, eerder ingevoerd door Makino voor classificatie in het geval van slechts twee klassen. In hoofdstuk 3 wordt een algoritme geformuleerd, dat een zogeheten quasimonotone beslisboom induceert. Ofschoon dergelijke bomen ook geen monotonie kunnen garanderen op uitbreidingen van de trainingsset, vertonen ze wel de zwakkere quasimonotonieeigenschap. Ook in dit hoofdstuk wordt alle theorie in formeel opzicht uiterst gedegen behandeld, inclusief correctheidsbewijzen. Tevens wordt het geheel met experimenten toegelicht.

AFBEELDING VAN BESLISBOOM OP GEWICHTSPARAMETERS

Het laatste hoofdstuk handelt indirect toch weer over beslisbomen. Volgens een door Park ontworpen methode kan men de beslisboom afbeelden op de gewichtsparameters van de onderste laag van een neuraal netwerk met één hidden layer. De auteur toont aan dat, in tegenstelling tot wat Park beweert, de classificatie van het neuraal netwerk niet geheel overeenkomt met die van de beslisboom. Wel zijn de beide classificaties bij benadering identiek. De kwaliteit van de benadering is instelbaar via parameters. De afbeelding van een beslissingsboom op een neuraal netwerk is mogelijk relevant voor het ontwerpen van een geschikte architectuur van het neurale netwerk.

BIVARIATE BESLISBOMEN

In hoofdstuk 4 worden bivariate beslisbomen besproken. De vorige hoofdstukken beschouwden slechts univariate bomen. Deze voeren per knoop een test op slechts één attribuut uit. Bivariate bomen kunnen in een knoop ook een test met twee variabelen uitvoeren. De bekendste in deze categorie zijn CART en OC1. Twee nieuwe algoritmen, BIT1 en BIT2 geheten, worden gepresenteerd. Evenals de twee oudere algoritmen testen de beide nieuwe slechts lineaire combinaties van attribuutwaarden. De rekentijd nodig voor het induceren van dergelijke beslisbomen is nauwelijks ongunstiger dan in het geval van univariate bomen. Overigens speelt in dit hoofdstuk het verschijnsel monotonie geen rol meer.

BNVKI newsletter

VIJFTIG-PLUSSER

Het proefschrift is fraai uitgevoerd en maakt een zeer leesbare indruk. Zoals boven reeds opgemerkt wordt voor de theorie steeds een gedegen formele 37

April 2000

verantwoording gegeven. Het overwegend theoretische karakter van het proefschrift heeft de auteur er niet van weerhouden uitvoerige experimentele uitkomsten te vermelden. De promovendus is zelf verbaasd over zijn eigen productiviteit. Hij is ruim vijf jaar geleden als vijftigplusser de onderzoekswereld binnengestapt. Het voorwoord opent met de volgende zin: If a fortune-teller would have told me “You will defend your PhD thesis before your 55th birthday” five years ago, I would have ridiculed her'. Dr. Potharst blijkt ongelijk gehad te hebben en geeft dit ruiterlijk toe. Op overtuigende wijze heeft hij aangetoond vijf jaar geleden een juiste stap gezet te hebben.

to choose a Ph.D. thesis title containing the word "learning". I am however excited to read Gianluca Bontempi's Ph.D. thesis and would like to say that the framework of "local learning", or more specifically, "lazy learning", advocated in Gianluca's work is on the right track! Yes, we have to be "local" and we have to be "lazy" because in most of the cases we have no choice! Actually, nothing is wrong with the "blind men" just mentioned. They have no choice because of their visual sensor failure and they have to approach the "elephant identification problem" by local information only. This may be a passive example. Let us consider a non-passive one. Suppose you want to cross the road, you do not need the traffic information of this road 50 miles away. Your decision to across the road is based on the local information. In our real world, the loop of "sensing-decision-control" is largely "local".

LOCAL LEARNING TECHNIQUES FOR MODELLING, PREDICTION AND CONTROL Thesis by Gianluca Bontempi Iridia ULB

THE WHOLE LOOP INTRODUCTION

Edwin de jong, VUB

In AI research, people are now facing the so-called complex system which is asked to perform a complex task in a complex environment - "CCC". Global thinking is restricted by our socio-technological limits such as man and computational powers etc. Most naturally, people may be asked the question "why global when a cheap local scheme is good enough?" Another issue in AI research is the incomplete or broken loop of "sensing-decision-control". We generally have two objectives for our activities: "realise the world" and (then) "change the world". It seems that most of the AI research objectives are just to "realise the world", i.e., only the "sensingdecision" part. Gianluca's work under the label of "Supervised Learning in AI", however, touched the whole loop of "sensing-decision-control". This distinguishes his work, which includes modelling, prediction and control using

Gianluca Bontempi is researcher at the Philips Research Laboratories in Eindhoven, The Netherlands. From 1997 to March 2000, he was a Marie Curie fellow at the IRIDIA lab of the Universite Libre de Bruxelles, headed by Hugues Bersini and Marco Dorigo. On December 22 1999, he defended his Ph.D. in Applied Sciences, for which he received ‘la plus grande distinction’ (grootste onderscheiding). His paper with Mauro Birattari received the Best Paper Award at the previous BNAIC. Dr. Yangquan Chen, senior engineer at Seagate R&D in Singapore and expert on learning control, provided the following review of Gianluca’s thesis. Thesis review by Yang Quan Chen

Singapore Science Park Design Centre Seagate Technology International Email: [email protected] "Learning" is a quite general concept in engineering and in science. It is more like an "elephant" surrounded by a group of "blind men". One has to be brave enough BNVKI newsletter

38

April 2000

procedure is required to choose the best one. Both a competitive and a co-operative approach were explored in the thesis. The competitive approach is based on a winner-takes-all strategy while the cooperative strategy is based on the theory of the combination of estimators. Clearly, this kind of local learning algorithm is lazy due to the fact that it defers the whole learning procedure to the moment when a prediction is required. This is referred to as a Lazy Learning (LL) method. From a number of supervised learning benchmarks presented in this thesis, we can see that the Lazy Learning algorithm is able to compete with global and local state-of-the-art approaches. It is worth repeating here that the Lazy Learning method also took part in the Third International Erudit competition on supervised learning where it was awarded as a runner-up among 21 participants.

local learning techniques, from other AI research. CONTRIBUTION

The major contribution of this thesis is first in this thesis itself. It is well organised with a systematic presentation of local learning techniques. The carefully selected appendices made this thesis self-contained and yet the main body of the thesis can lead the readers to the cutting edge of this research area. It has been actually a well prepared book, to my point of view, ready for publication as a research monograph. Technically, the main contributions of the thesis are three-fold: (1) supervised learning methods; (2) modelling and timeseries prediction; and (3) non-linear dynamic system control. The thesis presents a set of original local modelling techniques and their applications to a number of experimental problems in regression, time series prediction and adaptive control. It has been successfully shown that the local paradigm is not only a powerful idea to design learning algorithms, but also a fruitful way of reasoning in different domains of applications.

LAZY LEARNING

The thesis contributes to long-term timeseries prediction based on the iteration of a one-step-ahead predictor. Gianluca advanced a Lazy Learning approach with local validation and bandwidth selection which is performed on the basis of an iterated formulation of the PRESS crossvalidation criterion. This is an efficient way for assessing the generalisation performance of a multi-step predictor on a horizon longer than a single step, which is totally new according to existing local modelling literature. The effectiveness of the proposed LL prediction method has been demonstrated by the successful results in the prediction of the chaotic Mackey Glass time series, in the prediction of two time series from the Santa Fe competition, as well as in the International Competition on Time Series organised by the International Workshop on Advanced Black-box techniques for non-linear modelling in Leuven, Belgium. In the competition, two prediction results obtained by using the proposed Lazy Learning algorithm ranked second and fourth, respectively.

In local learning methods, the key problem is the bandwidth selection. Gianluca proposed a datadriven technique for bandwidth selection which, for each query point, (i) generates a number of local model candidates with different numbers of neighbours, (ii) validates them via cross-validation and finally (iii) selects or combines the most promising ones. The set of models taken into consideration are local linear models centred in the query point and fitted with an increasing number of neighbours. A spatial sequence of neighbours that are incrementally considered as making part of the local model centred in the query point are fed into the proposed spatial recursive least-squares algorithm. Once the candidate models have been generated, the PRESS statistic, an efficient local cross-validation procedure, is used to assess the generalisation ability of each of them. This validation does not involve any significant computational overload, since the PRESS statistic uses partial results returned by the recursive least squares algorithm. Once the models have been generated and assessed, a model selection BNVKI newsletter

NON – LINEAIR CONTROL

As for the contribution to non-linear control, Gianluca investigated three major 39

April 2000

multi-model control using the notion of "practical stability" instead of the global Lyapunov stability notion.

control schemes, viz. the forward and inverse model control (direct method), self-tuning regulators (indirect method), and multi-model linearization control. In the forward and inverse model control, Gianluca used the Lazy technique to get the forward and inverse models of the system which are then used to compute the control signal. The control algorithm also includes a one time-step horizon and is implemented as a gradient based optimisation algorithm where the Lazy Learning model computes the value and the gradient of the cost function to be minimised. This kind of "feedforwardfeedback" structure is very popular in control engineering practice. LL technique used in constructing a good "feedforward" control is a new and a promising attempt in control engineering practice. Including LL techniques in the identification or estimation processes in conventional adaptive control will rewrite the theory in a new paradigm. Gianluca presented a Lazy self-tuning approach without using any global linear model description as in conventional adaptive control. Instead, at each time-step the system dynamics is linearized in the neighbourhood of the current operating regime. It should be note that the adaptive feature of the Lazy model is due not to a sequential parameter estimation but simply to the database updating. In multi-model linearization control method, a local estimator is combined with a time-varying optimal control. This can cater for nonlinearity of a system over a wider range than conventional linearized quadratic regulators. Some simulation examples of identification and control of non-linear systems with the three Lazy Learning control approaches are provided in the thesis. Clearly, the Lazy Learning Control (LLC) problem opened in this thesis for the first time calls for more rigorous development in both theory and in applications. One of the potential developments is under the framework of BNVKI newsletter

To my perspective, the lazy concept systematically explored in this thesis should find its appealing position in engineering practice. Again, why "global" if "local" is okay? Or, act in this way, increase the region of locality if this "local" is not good enough. Gianluca's thesis is available from http://iridia.ulb.ac.be/~gbonte

Section Editor Richard Starmans

SIKS-DAY 2000, OCTOBER 13, AMSTERDAM On October 13, the School for Information and Knowledge Systems organizes its annual SIKS-day in Amsterdam. The precise location is not known yet, but this early announcement only serves to draw the attention of our members. The main aim of the event is to give SIKS- members participating in research groups all over the country- the opportunity to meet each other in an informal setting and to inform them about current developments and some new activities and plans for the coming year. This year a small scientific symposium will be organised at the SIKS-day, as well. Three guest speakers have agreed to participate:

Frank van Harmelen (VU) Arno Siebes (CWI) Carles Sierra (University of Barcelona)

40

April 2000

All members of our research school are invited to join the SIKS-day 2000. More details about location, program and registration will be given in the next edition of this newsletter.

Please inform her whether you are SIKS Ph.D. Student or not. After your registration you will receive the final program and other details about the courses as soon as possible. For further questions about the content of the course, please contact the co-ordinator of SIKS. Email: [email protected]

BASIC COURSES

The core of the program is a two-yearly cycle of eight obliged courses, that cover important paradigms, topics and trends in the field of information and knowledge systems. Six of these courses are organised on a regular basis (B1 to B6), the remaining two, B7 and B8 may contain miscellaneous subjects, current trends and topics. In the previous edition of this newsletter we informed you about the next course “Databases” (B3) and “Interactive Systems” (B6) that will be organised in Hotel Apeldoorn from June 5 till June 9. Recently the scientific advisory council of SIKS decided upon the course program for 2000 and 2001. In order to facilitate an adequate planning for our Ph.D.-students a brief schedule is given here:

SECTION KNOWLEDGE SYSTEMS IN LAW AND COMPUTER SCIENCE Section-Editor Radboud Winkels

VERSLAG WETENSCHAPPELIJK DEEL JURIX BIJEENKOMST 9 FEBRUARI 2000, UVA MODELLEREN VOOR CONCEPTUAL RETRIEVAL IN

CLIME

Presentatie: ‘Modelleren voor Conceptual Retrieval in CLIME’ door Alexander Boer en Rinke Hoekstra (Rechtsinformatica, Universiteit van Amsterdam)

December 2000

Electronic Commerce (B7) Internet Computing (B8)

Verslag door Luuk Matthijssen (Centrum voor Recht, Bestuur en Informatisering, Katholieke Universiteit Brabant) Vaak wordt de term information retrieval (informatie ontsluiting) gebruikt voor iets wat niet meer is dan document retrieval (document ontsluiting). Wanneer een geautomatiseerd systeem als antwoord op een informatieverzoek van een gebruiker alleen documenten presenteert, is dit slechts document retrieval. De gebruiker moet zelf de informatie-inhoud van de documenten achterhalen, beoordelen en toepassen. Er is, strikt genomen, pas sprake van information retrieval wanneer het geautomatiseerde systeem de gebruiker ook ondersteuning biedt bij het interpreteren en het toepassen van de

May/June 2001

Combinatory methods (B4) Intelligent systems (B5) December 2001

System modelling (B1) Knowledge modelling (B2) REGISTRATION

Registration for the courses Databases and Interactive Systems (June 5 till June 9) is still possible. Anyone who is interested in taking these courses, is kindly requested to contact Margreet van Soest for registration! You can reach her preferably by e-mail: [email protected] BNVKI newsletter

41

April 2000

bijeenkomst een aantal methoden en geautomatiseerde hulpmiddelen die binnen het CLIME project worden ontwikkeld om deze modelleringsinspanning te ondersteunen.

informatie die in de gevonden relevante documenten besloten ligt. Een groot obstakel voor het effectief realiseren van information retrieval is dat de informatie in documenten veelal in de vorm van natuurlijke taal is neergelegd en daarom niet automatisch kan worden gelezen en geïnterpreteerd. De doelstelling van het CLIME project (Computerized Legal Information Management and Explanation) is om daadwerkelijke information retrieval te realiseren voor databanken met juridische documenten.

CLASSIFICATION OF SHIPS

De juridische documenten waarbij de databank van de LIS in dit project wordt gevuld, zijn de “Rules and regulations for the classification of ships” van Bureau Veritas (die onder meer gebruikt worden voor verzekeringen en de toegang tot havens) en internationale scheepvaart verdragen over veiligheid (SOLAS) en maritieme milieubescherming (MARPOL). Deze juridische documenten omvatten duizenden pagina’s tekst en zijn beschikbaar in elektronische vorm. Typische vragen die de LIS op basis van de juridische informatie in deze documenten moet kunnen beantwoorden zijn: “Mag een bulk-vrachtschip ook passagiers vervoeren?” en “Hoeveel hoospompen moeten er minimaal aanwezig zijn op een vrachtschip?”. Om deze vragen te kunnen beantwoorden, moet de normatieve status van de omschreven casus worden beoordeeld. Daarvoor moeten de toepasselijke normen worden gekoppeld aan de casus omschrijving van de gebruikersvraag. Deze koppeling kan veelal niet direct gemaakt worden omdat de zoekvragen de casus maar zeer gedeeltelijk omschrijven en omdat de toepasselijke normen vaak veel abstracter geformuleerd zijn dan de concrete casus. Het is daarom nodig om behalve de abstracte rechtsregels ook “reële wereldkennis” te modelleren waarin wordt aangegeven hoe de abstracte concepten uit de rechtsregels zich verhouden tot de specifieke eigenschappen van de casus en waarmee de geïmpliceerde eigenschappen kunnen worden afgeleid waarmee de beperkte casusbeschrijving kan worden aangevuld. Voorbeelden van zulke reële wereldkennis zijn dat zowel de bemanning als passagiers van een schip natuurlijke personen zijn, dat een vrachtschip een vaartuig is en dat een hoospomp een onderdeel van een vaartuig is. In de modelleringstaal wordt de relevante kennis uit de juridische documenten beschreven in de vorm van regels. Regels beschrijven situaties (in termen van predikaties van concepten over objecten) en normen (die een uitspraak doen over het verboden of toegestaan zijn van situaties beschreven in casus). Verder hebben regels en concepten ook “metajuridische” attributen zoals ingangsdatum, einddatum, regelgevende autoriteit, betrokken partijen, etc.

CLIME

CLIME is een ESPRIT project waarin door de Universiteit van Amsterdam wordt samengewerkt met British Maritime Technologies (UK), University of Brighton (UK), Bureau Veritas (Frankrijk) en TXT (Italië). Het project als geheel heeft als doel om technieken te ontwikkelen waarmee via het Internet toegang kan worden geboden tot juridische informatie op het gebied van internationaal maritiemrecht. Het deelproject waarvoor de UvA verantwoordelijk is betreft het ontwerp van een Legal Information Server (LIS). De LIS moet op basis van vragen van gebruikers uit de verzameling documenten in een juridische databank de toepasselijke rechtsregels kunnen ophalen en presenteren. Daarbij wordt niet volstaan met het presenteren van juridische documenten maar worden ook de rechtsregels die in deze documenten zijn vervat, gepresenteerd en gebruikt voor redeneren over de juridische consequenties van een gebruikersvraag. Het resultaat is een systeem dat geautomatiseerd een juridische beoordeling kan geven van een casus die door een gebruiker in een vraag aan het systeem wordt omschreven. Het probleem van de geautomatiseerde interpretatie van natuurlijke taal wordt opgelost door de rechtsregels in de documenten van de juridische databank expliciet te modelleren en formeel te representeren. Op deze manier kunnen de rechtsregels uit de documenten automatisch worden gelezen en toegepast. Het nadeel van deze oplossing is dat er bij het opbouwen van de databank een fikse modelleringsinspanning moet worden geleverd waarbij alle rechtsregels die erin besloten liggen door menselijke experts moet worden geïnterpreteerd en in een formele taal moeten worden vastgelegd. Alexander Boer en Rinke Hoekstra presenteerden bij de JURIX BNVKI newsletter

PROBLEEM

42

April 2000

Legal Encoding Tools bieden ook functies voor onderhoud en beheer van kennisbestanden. Zo kunnen kennisbestanden worden gewijzigd en samengevoegd en kunnen ook nieuwe of veranderde documenten aan de databank worden toegevoegd.

Eén van de problemen waar men bij het modelleren van deze kennis op basis van de juridische documenten in de databank tegen aan loopt is dat een concept waar door verschillende regels aan wordt gerefereerd maar eenmalig moet worden gemodelleerd. Bij het modelleren van regels moet dus herkend worden wanneer een concept waaraan door een regel wordt gerefereerd hetzelfde is als een reeds gemodelleerd concept. Een ander probleem is dat relaties tussen normen expliciet moeten worden gemaakt. In principe is het mogelijk om bij conflicten tussen toepasselijke normen een oplossing af te leiden op basis van metanormen waarin wordt aangegeven welke norm in het onderhavige geval prevaleert. Omdat deze conflictresolutie een grote redeneercapaciteit vereist, is ervoor gekozen om de relaties tussen regels expliciet te modelleren waardoor gebruikersvragen sneller en met minder systeembelasting kunnen worden beantwoord. De relaties tussen regels worden weergeven in een uitzonderingshiërarchie waarbij het ‘lex specialis derogat legi generali’ principe wordt gehanteerd wat zegt dat specifieke regelgeving vóór algemene regelgeving gaat. De geautomatiseerde hulpmiddelen die voor het modelleren van de juridische kennis in het CLIME project worden ontwikkeld (Legal Encoding Tools), bieden functies die deze problemen voor de informatie-analist inzichtelijk en hanteerbaar moeten maken.

STATUS

De status van de modelleringsinspanning is dat momenteel ongeveer 10% van alle documenten in de databank is gemodelleerd. De onderzoekers verwachten een grote stijging in het modelleertempo omdat nieuw ingevoegde documenten steeds vaker zullen refereren aan reeds gemodelleerde concepten. Tot dusver heeft het UvA deel van het CLIME project al verschillende interessante wetenschappelijke resultaten opgeleverd waaronder verfijningen van juridische redeneermodellen en specificaties voor geautomatiseerde modelleringshulpmiddelen. Deze resultaten zijn gepresenteerd in wetenschappelijke artikelen op, onder meer, de congressen JURIX ’98 en ICAIL ’99. Meer informatie over het CLIME project is te vinden op de website http://www.bmtech.co.uk/clime/. JURIX AND PUBLIC COUNTER 2000

JURIX lecture (February 9, 2000 in Amsterdam) by Joël van de Meer (Impactive)

De tekst van een te modelleren stuk regelgeving kan in de editor van de Legal Encoding Tools (LET) worden ingeladen. Vervolgens kunnen de norm-elementen en concepten die door de informatie-analist worden herkend, worden gemarkeerd en ingevuld in een norm editor of een concept editor. In deze editors wordt tevens aangegeven wat de overige verplichte attributen zijn die moeten worden ingevuld. Daarbij worden ook overzichten getoond van de reeds bestaande normen en concepten. Zo kunnen de nieuwe regels op de juiste plaats in bestaande kennisbank worden ingevoegd waarbij ook de relaties tussen de normen en de concepten kunnen worden gespecificeerd. De BNVKI newsletter

Report by Ronald Leenes Faculty of Public Administration and Public Policy University of Twente, Enschede Currently, the public sector is organised along the lines of the various sectors that comprise its physical organisation. The administration on all levels of the state is organised in sectors. The Ministry of Social affairs is divided into sectors, each 43

April 2000

dealing with some particular aspect of social affairs. At the local level, the same departmentalisation and specialisation can be seen. While this specialisation may be efficient from the perspective of developing policy, the effect for the citizen as a client of public services is less beneficial. When a citizen encounters a problem with 'public' dimensions, she typically has to deal with various offices, desks and counters. For a client with a complex problem it is often difficult to see the forest for the trees. This can clearly be seen in the domain of building and housing. A client who wants to build an office in her garden needs all kinds of permits (building permit, logging permit, car entrance permit), and may be entitled to subsidies (sustainable building subsidy, energy saving subsidy). All these permits and subsidies are administered by different offices.

•

•

•

OVERHEIDSLOKET 2000

This problem of fragmentation from the perspective of the client needs to be addressed. This is exactly what the Dutch programme “Overheidsloket 2000” (Public Counter 2000) aims to accomplish. This programme aims at a 'tilting' of public service provision from product-centred to customer-oriented. Changing towards customer-oriented service provision is a reason in itself to change. But, there are more reasons to rethink public service provision. To name a few: • Quality improvement In almost any field there is a tendency towards quality improvement as witnessed by ISO 900x, Service level agreements, charter marks etc. • Competition Although public services are delivered by monopolists, there are tendencies to break or change the monopolies. In the welfare sector, for instance, insurance companies, together with banks are BNVKI newsletter

creating total-care products, competing with the traditional welfare institutions. On the other side of the spectrum there are voices to reinstate centralised passport provision, a task now performed by local communities. Technology Technology on the one hand enables service integration, and is a necessity for service integration on the other. Administering a multitude of services instead of a few products calls for ICT support. Efficiency Maintaining many counters as opposed to a few integrated service desks and integration of forms leads to greater efficiency. Effectiveness A major problem of many public policies is the limited target group reach. Studies have shown target group reach of about 30% in the field of housing subsidies. One of the factors that explain these low numbers is the lack of bureaucratic competency of the citizens and the fragmentation in service provision.

These problems are addressed in the “Overheidsloket 2000” programme initiated in 1996 by the Ministry of Internal Affairs. The key idea in this programme is integrated service provision on the basis of problem clusters from the client's perspective. These problem clusters are called 'demand patterns', a somewhat clumsy term because it suggests that it concerns manifest client demands. Instead, a demand pattern denotes the cluster of problems (and solutions) underlying a concrete demand. So, the demand pattern 'building' contains all aspects (and products) of building an object (such as a house). In the public sector this amounts to some 40 products and services! The reason why a demand pattern (in some instances: life event such as birth, passing away, 44

April 2000

In 1999 phase two of the programme was started (ending in 2003). The focus is now on the widespread implementation of three kinds of agencies: building and housing (under supervision of the Ministry of VROM), welfare (Ministry of VWS) and businesses (Ministry of EZ).

marriage) covers a deeper level than the question a client poses, is of course that clients often are not aware that certain products exist. Hence, they will not ask fore them. The demand pattern can be seen as a checklist that contains all possibly relevant products and services relating to a topic.

ICT plays an important role in meeting the goals of integrated service provision. All back-offices make use of IT systems. If part of the service provision is transferred to a front-office, this front-office needs to have access to the back-office IT systems. This calls for the construction of datawarehouses, mid-offices, networks etc. In the front-office there is a need for intelligent support. The front-office worker has to have (expert) knowledge on a much broader area than before. The typical frontoffice worker has to be a generalist instead of an expert, but at the same time the level of expertise has to be high (HBO level). A front-office worker may benefit from knowledge based systems and other intelligent support tools. These systems are to provide easy access to related products and services and have to support the integrated assessment of problems.

All products and services of a particular demand pattern are to be administered in a single front-office. Ideally, a client who wants to build an office for instance, only needs to visit the Building and housing window to obtain all permits and subsidies. The advantages of this approach are clear: no more pillar to post, filling in less forms, less redundancy in intake etc. “Overheidsloket 2000” focuses on the front-office. It builds on a separation between front-office, where all client contacts take place and where as many as possible services are actually provided, and the back-office where the experts reside, the more difficult products are handled and where policy is made. It is an ambitious programme in which all kinds of problems emerge: turf protection (different departments need to work together), lack of conceptual knowledge (what are relevant demand patterns), technical problems (how to support the front office worker), cultural problems (we are doing fine, so why change) etc. The way to tackle these problems is by means of doing pilots. This was the approach in the first phase of the project. In fifteen pilots local municipalities have tried to implement one-stop-government agencies on the domains: building-and-housing, welfare (elderly and disabled) and know-yourrights. The pilots have generated many insights in how to implement one-stopgovernment agencies.

A step further is the creation of virtual onestop-government agencies. Many Overheidsloket 2000 projects (e.g. www.enschede.nl/ole2000) aim at creating such a virtual counter. Here not only providing integration between services is a major problem, but also the interface with the client. In comparison with a highly skilled front-office worker, the average client is a lay person. The virtual office has to guide the user through the forest of public services by means of an intelligent intake. This is a long way from current practice. Many local communities currently offer information about their products and services, but they do so by letting the user choose from lists of products (e.g. www.tilburg.nl/stadhuis/balies/balie

PHASE 2

BNVKI newsletter

45

April 2000

index.htm). A more appropriate approach would be to first diagnose the client's problem and then focus on the relevant products and services. An example of this approach can be found on: http:// www.impactive. nl/itr/default. htm.

WIZWISE TECHNOLOGY (WWT)

Lezing van Jan van Lieshout, 9 februari 2000 te Amsterdam Jan van Lieshout, van oorsprong fysicus, is werkzaam bij de onderneming WizWise Technology. Als fysicus heeft Van Lieshout zich altijd al beziggehouden met vragen zoals: ‘Wat betekent een formule in de werkelijkheid?', 'Hoe kom je daartoe?' en 'wat is de impact van de toepassing van modellen?'. Ook binnen WizWise richten hij en de andere medewerkers zich op deze en andere modelleringsvragen.

INTRODUCING KNOWLEDGE BASED SYSTEMS

Knowledge based systems can be used to diagnose a client's needs. They can also be used at a lower level. Individual products and services, such as a building permit, are often legal products. Their application is based on rules and regulations. KBS technology can be used to build components (kbs agents) that can handle the processing of these products. An interesting development is the construction of a virtual one-stop government counter consisting of an intake module, a mechanism to co-ordinate the proper kbsagents and a number of task specific agents. (Within Jurix we try to get a research project along these lines started.)

Van Lieshout trachtte met zijn lezing antwoord te verschaffen omtrent de volgende vragen: ‘Wat probeert WWT te bewerkstelligen?’, ‘Wat is kennismanagement?’, ’Waar staat WWT?’ en ‘Wat is er specifiek voor juridische domeinen?’. WAT PROBEERT WWT TE BEWERKSTELLIGEN?

Much remains to be done before we can truly benefit from integrated (virtual) onestop-government counters. It is clear AI (and Law) has a role to play in this field.

WizWise is een organisatie die zich, in het kort gezegd, richt op het ondersteunen van organisaties bij het toepassen van Kennismanagement (KM). Deze ondersteuning uit zich onder meer in het geven van advies omtrent diverse benaderingswijzen van KM, het hanteren van KM-methodes, het opzetten en inrichten van architecturen en het ondersteunen met systemen en gereedschappen. WizWise richt zich ook op het onderzoek naar en de toepassing van kennismodelleermethoden zoals onder andere KADS en UML, webtechnologieën, open standaards zoals o.a. Java, Corba, HTML en XML, en overige ontwikkelingen op het gebied van kennisgebaseerde systemen en methoden zoals agentsystemen. WizWise doet dit niet alleen maar in verschillende samenwerkingsverbanden met specialisten

KNOWLEDGE MANAGEMENT

(IN EEN JURIDISCHE CONTEXT) Verslag van de JURIX lezing van 9 februari 2000 te Amsterdam van Jan van Lieshout Verslag door Martin Apistola Sector Informatica

Haagse Hogeschool

BNVKI newsletter

46

April 2000

ondersteuning daarvan met behulp van Informatietechnologie (IT), kunnen we onderscheid maken in proces georiënteerde Informatietechnologische (IT) oplossingen en (informatie)product georiënteerde IT oplossingen. De primaire taak van de procesgeoriënteerde ondersteuning, dat begin jaren negentig opkwam, is het ondersteunen van (groeps) communicatie en samenwerking. Hiertoe bedient zij zich van kennisbronnen zoals onder andere email berichten, afspraak notities, agenda’s en workflow management, ondersteund door o.a groupware en workflow management systemen. Product georiënteerd, dat sinds eind jaren negentig in opkomst is geraakt, betekent het actief en gericht bezorgen van relevante informatie aan een “kennis intensieve taak”. Hiertoe maakt zij gebruik van kennisbronnen zoals onder andere proceshandboeken, geleerde lessen en ontologieën, ondersteund door o.a. databases en document management systemen.

zoals onder andere de Erasmus Universiteit Rotterdam, AA en Syntens. WAT IS KENNISMANAGEMENT?

Een organisatie bestaat uit een aantal essentiële componenten. Zo zijn er de mensen en (hun) kennis die van belang zijn voor het voortbestaan van de organisatie. Het juist beheren van, en omgaan met deze componenten is dan ook cruciaal. Natuurlijk ook gezien vanuit het perspectief van de huidige informatiemaatschappij waarbinnen de hoeveelheid en diversiteit aan informatie groteske vormen heeft aangenomen en in extreme gevallen zelfs tot een stressfactor kan leiden. Maar ook om als onderneming in de huidige informatiemaatschappij voor te lopen op de concurrentie zal de onderneming haar kennis op een zo’n effectieve en efficiënt mogelijke wijze dienen te beheren. Een veelgebruikte term voor het beheren van kennis is Knowledge(kennis)management (KM). Dit beheer kan echter niet juist worden uitgevoerd zonder aandacht voor de menselijke aspecten binnen organisaties zoals cultuur, relaties, opleidingen, etc. WizWise zelf ziet KM als middel “(..) to manage and support knowledge intensive work (KIW)”. De vraag die dan naar boven komt is wat er dan onder een kennis intensieve werker en zijn werk verstaan wordt. Wat in ieder geval opvalt aan een KIW is dat hij te maken heeft met ‘’wicked’’ (steeds verschillende soorten) problemen, waarbij de ondersteuning van een geautomatiseerd gereedschap noodzakelijk is. De informatiescope van de KIW valt zeg maar buiten de scope van de traditionele informatievoorziening. Als voorbeelden van kennisintensief werk kunnen tenslotte de acquisitie, creatie en verspreiding van kennis genoemd worden. Volgens Van Lieshout kunnen we KM tevens vanuit drie perspectieven opvatten, namelijk het proces-, product- en organisatorisch perspectief. Voor de KM BNVKI newsletter

WAAR STAAT WWT?

Binnen de markt voor de ontwikkeling en toepassing van KM oplossingen blijken er tegenwoordig verschillende invullingen van het begrip kennismanagement te zijn. Zo bestaan er bij de verschillende aanbieders van KM verschillende georiënteerde adviezen en oplossingen, eigen aanpakken en raamwerken die gehanteerd worden om de gebruikte KM begrippen in op te nemen. Een tweetal specifieke voorbeelden van verschillende KM-aanpakken uit de praktijk zijn de aanpak volgens Knownet en Ernst & Young. Knownet is een project dat nastreeft om een KM-methode/raamwerk, een ondersteunende architectuur en gereedschappen aansluitend op de betreffende KM-methode op te leveren. De KM-aanpak van Ernst & Young richt zich specifiek op mensen, processen en infrastructuren. Het blijkt volgens Van 47

April 2000

Lieshout ook dat de meerderheid van de KM-methodes zich enkel op bepaalde organisatiecomponenten (bijv. alleen technologisch of sociaal) richt. Het komt nog maar weinig voor dat er sprake is van een integratie van KM opvattingen. Daarnaast blijken er ambiguïteiten in de KM-terminologie te bestaan waarmee KMopvattingen beschreven worden. Dit zijn een aantal conclusies getrokken na het onderzoek van WWT in diverse projecten naar KM-methodes en raamwerken. De diverse methodes en raamwerken werden door WWT vergeleken om overeenkomstige sterke en zwakke punten te vinden. Voor de analyse van KMraamwerken kan er gebruik gemaakt worden van de WizWise FIA-analyse methode. Deze methode wordt gebruikt om zicht te krijgen op hoe KM geschiedt en mogelijkerwijs verbeterd kan worden in een organisatie. Voor met name de analyse en modellering van het "manipuleren" van kennis en informatie, wordt door WWT de Kennis Manipulation Diagram (KMD) methode gebruikt. Voor het toepassen van KM maakt WWT tevens gebruik van het Starship Enterprise Organisational Memory (SEOM)-model. SEOM is een conceptueel model waarin de interactie tussen de processen, de aanwezige kennis en organisatorische opvattingen wordt afgebeeld. Meer specifiek wordt het model onderverdeeld in organisatorische processen (o.a. beheerprocessen en productieprocessen), een Knowledgeable Information Broker (de hoofdschakel (‘makelaar’) tussen de ondersteuning van de processen, de uitvoerders daarvan en de bronnen waaruit de kennis voor deze ondersteuning geput en opgeslagen kan worden) en organisatorische bronnen (o.a. databanken en het World Wide Web).

toch moeten samenwerken. Zij hebben elk eigen verantwoordelijkheden, maar moeten wel elkaars informatie kunnen gebruiken. Het is dus voor deze groep onontbeerlijk gebruik te maken van open standaards. De problemen die hier met name om de hoek komen kijken zijn problemen omtrent de beveiliging van bijvoorbeeld gegevens. Voor de juridische domeinen zijn er mogelijkheden zoals een zogenaamd ‘’portal’’ systeem waardoor elke (juridische) gebruiker zijn informatie zowel ophaalt als opslaat, geassisteerd door alle functionaliteiten die er in het ‘’portal” beschikbaar zijn. Om te komen tot een KM (portal) systeem voor een (juridische)organisatie, zijn er twee opties te bedenken: (1) Alle vraag naar, en opslag van informatie door de gebruikers geschiedt via een ‘point of access’ (poort) zodat alles constant bijgehouden en geupdate kan worden, (2) we bezien de ‘wereld’ als verschillende organisaties, desnoods elk met hun ‘’privé-portal’’ achtig KM-systeem, waarbij de verschillende KM-systemen elkaar op de hoogte moeten houden/ondervragen over elkaars informatie/kennis. Volgens Van Lieshout zou voor juridische organisaties deze laatste optie het beter passen dan de eerste. CONCLUSIE

Kennismanagement is en blijft een moeilijk te begrijpen en te hanteren begrip. Dit bleek niet alleen uit het feit dat er gedurende de lezing nog gediscussieerd werd over het feit hoe ‘’kennis’’ geïnterpreteerd zou moeten worden, en uit het feit dat er tegenwoordig zelfs al ondernemingen blijken te zijn die zich in KM toepassing en onderzoek specialiseren, maar ook uit het feit dat er naast het begrip kennis er nog zo ontzettend veel andere KM-begrippen, KM-methoden en KMmodellen gebruikt worden. De presentatie van Van Lieshout bevatte in mijn ogen helaas een teveel aan uit te leggen KM-

WAT IS ER SPECIFIEK VOOR JURIDISCHE DOMEINEN?

Wanneer we kijken naar het recht, dan zien we vrij veel zelfstandige organisaties die BNVKI newsletter

48

April 2000

thought, concepts, communication systems, and the general development of intelligence in humans and other species, but moreover this investigation can be (and is) actively pursued with exactly the kind of computer models that are used in artificial intelligence research. In this respect the conference is relevant for AIresearchers that are interested in investigating the nature of human intelligence.

begrippen, KM-methoden en KMmodellen waardoor er meer verwarring geschept werd dan dat het KM verduidelijkte. De eerste versie van dit verslag is dan ook menigmaal aangepast en ingekort (een uitgebreidere versie is overigens op mijn internetpagina te zien). Een tweede punt van kritiek tot slot is dat de titel van de presentatie, namelijk ‘Knowledge Management (in een juridische context)’, in mijn optiek valse verwachtingen schiep, daar er hoogstens een tweetal korte ‘juridische KM’ opmerking over juridisch kennisgebruik en portaalsystemen werden gemaakt en er verder nauwelijks werd ingegaan op de juridische context.

This year’s conference consisted of 6 invited lectures, 58 oral presentations, 18 poster presentations and 6 discussion sessions. In order to accommodate so much in three days, three parallel tracks of talks were needed, while some discussion sessions were also held in parallel. Approximately 170 people participated.

END SECTION KNOWLEDGE IN LAW AND COMPUTER SCIENCE

THEME

THE EVOLUTION OF LANGUAGE CONFERENCE

The informal theme of the conference seemed to be the cognitive capacities of non-human primates. Four of the six invited lectures were on this subject. Talks by Sue Savage-Rumbaugh, Frans de Waal, Herb Terrace and Michael Tomasello convincingly showed that non-human primates have cognitive and social skills that are far better than is often assumed. The consensus seemed to be that higher primates have all of the cognitive and social skills that were once supposed to separate humans from other animals, such as ability to manipulate (sequences of) symbols, theory of mind, tool use etc. This, I think, was very interesting, not just because it places human intelligence in perspective, but also because it influences the notion of what human intelligence actually is; whether it is qualitatively different from primate intelligence or just quantitatively.

Bart de Boer Vrije Universiteit Brussel

From April 3–6 the third Evolution of language conference was held at the École Normale Supérieure des Télécommunications in Paris. The conference, which was organised by Jean Aitchison, Jean-Louis Dessalles, Jim Hurford, Chris Knight and Luc Steels, is a biannual event that brings together researchers from many different disciplines, such as linguistics, psychology, ethology, palaeontology, primatology, anthropology and biology, but also from artificial intelligence. Its aim is to provide a forum for an exchange of ideas on how language could have evolved in the human species.

AI RESEARCH

But of more direct relevance for the research into artificial intelligence were the presentations of work on computer models of the emergence of language. In this area of research the Belgian and Dutch participants of the conference were very well represented. Luc Steels, Joris Van Looveren, Guy de Pauw, Walter Daelemans, Jelle Zuidema, Paulien Hogeweg as well as the author himself presented posters and talks on this subject. In total there were two posters and about ten presentations

The question of how language can have evolved is, I think a relevant one for research into artificial intelligence. Not only does its investigation shed light on matters such as (among others) symbolic BNVKI newsletter

49

April 2000

that used computer simulations. Unfortunately it was impossible to attend all of them, because some of them were parallel to each other. Some themes addressed in simulation work were the relative roles of evolution (or innate properties) versus learning, (and the Baldwin effect) the role of population dynamics on the development of language, the role of self-organisation and the influence of the properties of a learning mechanism on the kind of language that can be learned. All of these phenomena are extremely hard to study in real human languages and populations, but relatively easily with computer simulations. This sort of work is related to the field of artificial life and it was therefore not surprising to see quite a few artificial life researchers on the conference.

TOWARDS AN ONLINE DISTRIBUTION STRUCTURE? Kees Jonkheer

Senior consultant NetlinQ Groep N.V. PREFACE

For years, Intelligent agents have been considered a great promise in e-business. Applications of Web intelligence are software programs that execute functions on behalf of its user on the Internet. These actions are search, compare, learn, negotiate or collaborate. Intelligent agents combine at least a few of these functions. Also in the Netherlands agents are finding themselves in an ever increasing attention. An example of many, are the products of agent builders Tryllian and Smarthaven. Tryllian recently launched the Gossipagent and is successful with it up until the United States where huge interest exist for commercial application of the concept. The ambitions of Trillion are high: setting a world standard for an agent platform. The fact that Stichting NLnet and the Vrije Universiteit of Amsterdam recently obtained a ten-year budget for an ominous research project aimed at application of intelligent agents in combination with privacy-issues and security is an indicative example of the interest and the importance that is attributed to agent technology.

WHAT COMES OUT IS WHAT YOU PUT IN

It was also interesting to observe the reaction of researchers from other disciplines on the modelling work. These range from extreme enthusiasm through respectful disinterest to disdain. This became especially clear in the discussion session on modelling. Although most of the people present where sympathetic towards the modelling work (a selection effect that occurred because there was a parallel discussion session on the aquatic ape theory of the origin of language) an important question that arose was what the contribution of the use of simulations is. Some people seemed to have the idea that “what comes out is what you put in” or in other words that the determinism of the computer simulations is the same as complete predictability, whereas much of the interesting behaviour of the simulations is emergent and therefore hard to predict from the algorithm directly. Another sentiment seemed to be that the computer models, in order to be useful, should predict actual properties of human languages. Questions like these are also relevant for AI-researchers trying to sell and explain their work outside their discipline. EVALUATION

I enjoyed the evolution of language conference very much. The best aspect of the conference is that it is interdisciplinary and that I therefore learn a lot about the different subjects that might be relevant for my own research and which I would not easily study myself. Also the talks are usually slightly adapted to the interdisciplinary and therefore nonexpert audience, so that they are easier to follow. It has not yet been decided where and when the next conference will be held, but I think it is worthwhile for AI-researchers interested in the origins of intelligence to keep this conference in mind.

AI IN BUSINESS BNVKI newsletter

SACRIFICE GAP

The emergence of massive e-commerce (buying via the Internet), the ever-growing need for businesses to improve customer contact and mobile internet mean a definitive break-through for web intelligence applications and agent technology in particular. Moreover, the 50

April 2000

Nasdaq-stockprice of the company has risen a thousand percent since the IPO in July 1998.

renewed attention for agents corresponds with the tendency of the quest for a fitting interface between the user and the WWW. Within 5 years it may be hopelessly outdated to surf on the WWW. Will every business own its website, visited by endless stream of searching and clicking customers? In e-business there appears to be a huge gap in the value added of online fulfilment. The aim should be to minimalize this so-called sacrifice gap: the difference between what the customer wants and what it eventually gets. Agents will offer a solution to this problem. It will become a discriminating factor in the evolution of an online distribution structure.

CUSTOMER CONTACT

Until now websites are strictly self-service environments which do not offer the same kind of benefits as offline active, skilful and personal customer care. Companies such as General Motors, Mercedes-Benz, Tower Records and Egghead Software try to incorporate the quality of traditional customer care by integrating call centre service on their websites. Call centre personnel can answer questions by using the telephone or internet chat connections. Although useful for customers, this approach is very expensive, labour intensive and of course scalability is a problem. So this calls for different, more automated solutions. More and more, online customers can be offered the right supply on the right moment or through new (combinations of) services. This new interactive approach leads to new forms of value creation. An important form of added value for the customer is the one-to-one approach: the supplier tries to identify the individual customer needs and offers fully individual propositions. To offer an optimal online individual servicing, occasionally companies are offering additional software for callcentre personnel, or guaranteed service for the most important customers. Othe applications are customer support by several forms of mass customisation, for example: expert systems, collaborative filtering systems, profile-based information push and question-answer agents. The number of online Customer Relations Management Tools is growing in a high pace, with a central focus the introduction of an interface that should make it easier for customers to execute actions on the Webster, such as navigation. Examples of companies that offer such products are Peoplesupport, Neuromedia, Chatrep and Artificial Life. This results in all kinds of sales reps, some of which are real time approachable (‘push-to-talk button’).

MASSIVE E-COMMERCE

Massive e-commerce stimulates the need and possibilities of applications of web intelligence and more specific, of agent technology. This is illustrated by the following example. Because of the massive participation in e-commerce, in the United States the demography of online buyers has changed dramatically over the last two years. Initially, it was the higher-educated male between 20 and 35 years that bought on the Net, now online buyers are found in every segment of society. Did the first group buy out of seeking convenience and curiosity, now price is the driving factor in online buying. That explains why shopping bots and price comparing sites with botfunctionality have become very popular over the last two years. Softwarebuilder Autonomy sees its existence also largely based on massive Internet-use. ‘Autonomy considers itself as ‘the Oracle of unstructured data’. Two main forces are push factors for the usability of their products: personalization of service and the information-explosion in text-form (WIRED 8.02, February 2000). It uses socalled Concept Agents that on the basis of context analysis retrieve information out of a huge amount of unstructured data. The BNVKI newsletter

It is obvious that developers of agent technology see an important role of their products in this area. Artificial Life sells one of its web intelligent ‘agentlike’ products, the so-called Web Guide to ecommerce retailers, financial services, consultancy bureaus and content providers. PriceWaterhouseCoopers uses Artificial Life’s ‘Personal Tutor’ for internal knowledge management. The within months expected Dutch portal for home improvement MyHome, on which 51

April 2000

200.000 products will be sold, introduces a digital passport for users. A logical next step will be to connect autonomous action functionality to the digital passport, so developing into the functionality of intelligent agents.

products. Because of this, functions in the physical distribution chain move upwards. The actual production will be done by the former main supplier. The function of the main supplier will be executed by former supplier and jobbers, etc.

MOBILE INTERNET

In such a concept there has to be a real-time customer approach (such as a real time chat partner, e.g., an online product advisor), that can test, incorporate preferences, etc. to eventually offer the customer an individually customised product. The ‘customising’ of a product becomes increasingly important in business-to-consumer relations, but is probably even more important in business-tobusiness relations. There, the customer itself is a producer too, which uses the purchased product in its own production. Applicability of the product is thus crucial.

In report published in October 1999, research institute OVUM expects intelligent agents to be the primary interface between the WWW and the user. The market for personal assistant services and intelligent agents for mobile phones will increase in 2005 up to 26 billion dollar. With a combination between WAPenabled browsers and personal agents, a new personal interface will evolve, that forms a personal mobile portal on the Internet. The report even states that agents will be the sole interface between the user and the WWW. Instead of customers entering the Internet to buy products, agents will login on the Internet to order and pay the products.

The physical distribution chain recognises the importance the communicative nearness to a customer, In an ‘online distribution chain’ such a role can be executed by intelligent agents. Partly, because it is impossible for companies to maintain optimal customer relations aside of their core businesses and partly because intelligent agents are capable of integral efficient information supply and execute all kinds of tasks. AN ONLINE DISTRIBUTION STRUCTURE

The combination of intelligent agents and mobile devices is currently put into practise by the United States stock-notated company AlphaServe.com that has developed intelligent agents for the Palm VII. Making use of the self-developed Network Query Language, the intelligent agents can execute several tasks such as the filling-in of forms, gathering of information and making transactions. Also, for example, a company such as Autonomy has made the connection with mobile devices by their so-called i-WAP technology.

Distance and added value are discriminating factors in the physical distribution structure. An example: a farmer produces milk, the milk plant pasteurises and packs the milk and supermarkets present the milk to the customers. To buy milk at the source, it would take the customer too much time and trouble to obtain the product and it will be deprived of added value (pasteurisation and packing). Thus, the distribution chain takes care of delivering the product as quick and convenient as possible to the customer.

Like in the physical distribution structure, products need to be ‘ready’ for consumption for online products too. In this case, discriminators are information and added value. Distance (in the physical world) and information (in the virtual worlds) are comparable entities. Distance could be overcome if it really has to be, just as information can be found if it really is necessary. But it just can be too inefficient and inconvenient in terms of money and time, which makes it impossible to obtain the required product or service. An online situation is

CUSTOMERS OFFLINE ARE IMPORTANT TOO

Customers are more and more committed in the production process. A relatively new example of such concepts is Demand Chain Management, what is considered as a form of Supply Chain Management. But, this time the producer will have a close relationship with the customer in terms of attention and client information, that it fully focuses on the development of individually-customised BNVKI newsletter

52

April 2000

industry, by which upon the basis technique a Knowledge Base is connected with branch specific data or knowledge.

considered to be largely based on information distribution, with a focus on buying decision. Also a customer is considered to be rational with which trendskips in preferences or fun shopping are ignored.

In this fashion, intelligent agents will play a decisive role in the evolution of an online distribution structure. The chains in the physical distribution structure of producer, wholesale, supermarket and restaurant, are analogues to the chains in the the online distribution structure, such as the company site, portals and intelligent agents. In this structure agents are the pervasive interface between the user and the Internet. Such agents are personal alter ego’s or representations of companies. As agents focus on certain specialisation’s, they perform more independent role. In the online distribution structure they really might evolve in brands themselves by creating an individual ‘identity’ (‘by that agent, I get quality for a bit of a high price’)

FUTURE

The functionality of intelligent agents depends not only of the advancement of the technical state-of-the-art but also on developments that enlarge the functionality of current intelligent agents on the web. Greater functionality will mainly show in larger opportunities for comparing, learning, etc. as well as the possibility to make transactions online. The report The New Economics of Transactions: Evolution of Unique e-business Internet Market Spaces published in May 1999 by Deloitte Consulting predicts a new fundamental transactions platform on the Internet. Six forces are held responsible for that. The three forces of relevance in the present context are the increasing role of intelligent agents, the evolving phenomenon of the electronic wallet, by which the difference between money and software will disappear, and the emergence of XML and XSL that replace HTML to make more efficiently distinctions between meta data and display data. The latter distinction is crucial for the autonomous functionality of intelligent agents, because they need metadata to function properly.

In summary it is plausible that under the influence of massive e-commerce, the crucial role of customer relations and mobile internet, it is just a matter of time before intelligent agents play a major, if not a pervasive role in e-business. The first commercial applications are entering the market just now.

What will be the application of intelligent agents in the near future? It will fill up the ‘open space’ in the possibility for supplying online added value. The main question here is how this interface will look like. And how it bridges the gap between perceived value and expected value of the customer. It is clear that such an interface should be personal, accurate and exhaustive building up an exclusive traction of trust. That role will be fulfilled by intelligent agents, in the function of a personal agent, for example, as a ‘personal portfolio manager’. As an aside, there will be agents that execute tasks on specific ‘skills’ for example finance or the purchase of consumer goods. A-Life recently produced a Smart Bot for the insurance BNVKI newsletter

CALL FOR PARTICIPATION

PLAYSTATION MEETS ORGANISM: EMULATION IN A CULTURAL CONTEXT 1-3 SEPTEMBER 2000, AMSTERDAM In association with Mediamatic Foundation, Amsterdam. The notion of emulation has recently 53

April 2000

gained new actuality, for example in game design, artificial life, sociology, biology and culture. In this seminar the meaning of emulation in these different discourses will be discussed, while exploring emulators and practical uses of emulation. The first day will focus on popular discourses in biology and artificial life, the second on emulation in the context of art, specifically new media art, and for the third day computer games and game design will be the central issue. Apart from the intellectual debate, participants will be able to revitalize themselves in real floating tanks and to experiment with computer games. Among the lecturers are George Dyson, Lev Manovich, Hiroshi Masuyama, Tom McCarthy, Trevor Pinch and Tom Ray. The seminar is intended for free-floating academics and joy-stick intellectuals. More detailed information is available at www.amsu.edu, or by e-mail at [email protected]. The Amsterdam-Maastricht Summer University P.O. Box 53066 1007 RB Amsterdam The Netherlands tel: +31 (0)20 6200225 fax: +31 (0)20 6249368 email: [email protected] web: www.amsu.edu

BNVKI newsletter

54

April 2000

August 6-18, 2000 12th European Summer School in Logic, Language and Information. University of Birmingham. Information: http://www.bham.ac.uk/~esslli/.

CONFERENCES, SYMPOSIA WORKSHOPS

August 21-22, 2000 The second International Cognitive Robotics Workshop held in conjunction with ECAI-2000 in Berlin, Germany. Information: http://www-15.informatik.nwth-aachen

Below, the reader finds a list of conferences and web sites or email addresses for further information. A more extensive list of conferences can be found in the Calendar 2000, as published in AI Communication and in the SIGART Newsletter.

September 3-8, 2000 International Summer School on Computational Logic. Maratea Italy.Information: http://www.cs.unimaas.nl /~etalle/school2000/index.html

May 8-10, 2000 Euromedia 2000, Antwerpen. Information: http://hobbes.rug.ac.be/~scs/conf/euromd2000/

September 5, 2000 DEXA 2000, International Workshop on Legal Information Systems and Applications LISA 2000 Information: http://www.univie.ac.at/RI/lisa2000.htm

May 13-16, 2000 An international conference on Artificial Neural Networks In Medicine And Biology. Göteborg University, Sweden Information: http://www.phil.gu.se/annimab.html

September 11-15, 2000 Automated Software Engineering – ASE’00 15th IEEE International Conference, Grenoble, France. More information: http://sigart.acm.org/Conferences/ase

May 15-19, 2000 The first Monet Summerschool on Model-Based Systems and Qualitative Reasoning (MBS/QR). University Residential Centre of Bertinoro (Italy). More information: http://monet.aber.ac.uk/summer_school/announcement.sh tml

September 18-22, 2000 Ninth International Workshop on Foundations of Models and Languages for Data and Objects, Schloss Dagstuhl, Germany. Information: http://www.cs.utwente.nl/~demm2000/

May 17-19, 2000 ICM, Information Systems for Enhanced Public Safety and Security, International Congress Centre, Munich, Germany Information: http://www.eurocomm.org/2000

September 20-23, 2000 Aimsa 2000, the Ninth International Conference on Artificial Intelligence: Methodology, Systems, Applications. Varna, Bulgaria. Information: http://www.inf.bas.bg/AIMSA2000

May 23-26, 2000 Second International ICSC Symposium, Neural Computation / NC’2000, Technical University of Berlin, Germany. Information: http//www.icsc.ab.ca/nc2000.htm

October 2-6, 2000 12th International Conference on Knowledge Engineering and Knowledge Management, Juan-Les-Pins, French Riviera. Information: http://www.inria.fr/acacia/ekaw2000

May 28-June 2, 2000 First European Workshop on RoboCup. Information: http://www.cs.uu.nl/people/wiebe/EuRoboCup2000/

October 27-29, 2000 Learning 2000 Workshop, University Carlos III of Madrid. More information: http://www. learn00.tsc.uc3m.es/

June 13-17, 2000 DNA6, Six International Meeting on DNA Based Computers. Leiden, The Netherlands Information: http://www.Icnc.nl/dna6/

December 12-15, 2000 Intelligent Systems and Applications (ISA’2000) University of Wollongong (near Sydney), Australia Information: http://www.icsc.ab.ca/isa2000.htm

June 19-21, 2000 COIL 2000, Chios, Greece. Symposium on Computational Intelligence and Learning. Information: http://www.dcs.napier.ac.uk/coil/.

December 14-15, 2000 JURIX 2000: Thirteenth International Conference on Legal Knowledge-Based Systems, Enschede, the Netherlands. Information: http://www.jurix.nl/jurix2000.html

June 26 – 30, 2000 Petri Nets 2000. 21st International Conference on Application and Theory of Petri Nets. Aarhus, Denmark Information: http://www.daimi.au.dk/pn2000

BNVKI newsletter

55

April 2000

Ir. E.D. de Jong Vrije Universiteit Brussel, AI Lab Pleinlaan 2, B-1050 Brussel, Belgium Tel.: +32 (0)2 6293713. E-mail: [email protected]

MAIL ADRESSES BOARD MEMBERS BNVKI Prof.dr. J. N. Kok Wiskunde en Natuurwetenschappen, Dept. of Computer Science Universiteit Leiden, Niels Bohrweg 1, 2333 CA Leiden Tel: (071) 5277057. E-mail: [email protected]

Dr. A. van den Bosch Katholieke Universiteit Brabant, Taal- en Literatuurwetenschap, Postbus 90153, 5000 LE Tilburg Tel.: (013) 4360911. E-mail: [email protected]

Dr. Y.H. Tan EURIDIS, Erasmus Universiteit Rotterdam Postbus 1738, 3000 DR Rotterdam Tel.: (010) 4082255. E-mail: [email protected]

Dr. R.J.C.M. Starmans Coordinator Research school SIKS, P.O. Box 80089, 3508 TB, Utrecht Tel.: (030) 2534083/1454. E-mail: [email protected]

Dr. E.O. Postma Universiteit Maastricht, IKAT Postbus 616, 6200 MD Maastricht Tel.: (043) 3883493. E-mail: [email protected]

Dr. B. de Boer Vrije Universiteit Brussel, AI Lab Pleinlaan 2, B-1050 Brussel, Belgium Tel.: +32 (0)2 6293703. E-mail: [email protected]

Dr. R. Verbrugge Rijksuniversiteit Groningen, Cognitive Science and Engineering Grote Kruisstraat 2/1, 9712 TS Groningen. Tel.: (050) 3636334. E-mail: [email protected]

HOW TO SUBSCRIBE

Dr. W. van der Hoek Universiteit Utrecht, Department of Computer Science, P.O. Box 80089, 3508 TB Utrecht Tel.: (030) 2533599. E-mail: [email protected]

The BNVKI/AIABN Newsletter is a direct benefit of membership in the BNVKI/AIABN. Membership dues are Fl. 75,- or BF 1.400 for regular members; NLG 50,- or BF 900 for doctoral students (AIO's); and NLG 40,- or BF 700 for students. In addition, members will receive two issues of the European journal AI Communications. The newsletter appears bimonthly and contains information about conferences, research projects, job opportunities, funding opportunities, etc., provided enough information is supplied. Therefore, all members are encouraged to send news and items they consider worthwhile to the editorial office of the BNVKI/AIABN newsletter. Subscription is done by payment of the membership due to RABO-Bank no. 11.66.34.200 or Postbank no. 3102697 for the Netherlands, or KBC Bank & Verzekering Veldwezelt No. 457-6423559-31, 20 Carabinierslaan 104, Belgie. In both cases, specify BNVKI/AIABN in Maastricht as the recipient, and please do not forget to mention your name and address. Sending of the BNVKI/AIABN newsletter will only commence after your payment has been received. If you wish to conclude your membership, please send a written notification to the editorial office before December 1 1999.

Dr. L. DeHaspe Katholieke Universiteit Leuven Dept. of Computer Science, Clestijnenlaan 200A,, B-3001 Heverlee, België Email:ldh@[email protected] G. Beijer BOLESIAN BV, Steenovenweg 19, 5708 HN Helmond Tel.: (0492) 502525. E-mail: [email protected] Dr. W. Daelemans Katholieke Universiteit Brabant, Vakgroep TaalLiteratuurwetenschap, Postbus 90153, 5000 LE Tilburg. Tel.: (013) 4663070. E-mail: [email protected]

en

EDITORS BNVKI newsletter COPY Dr. E.O. Postma (editor in chief) (See addresses Board Members)

The editorial board welcomes product announcements, book reviews, product reviews, overviews of AI research in business, and interviews. Contributions stating controversial opinions or otherwise stimulating discussions, are higly encouraged.

Prof. dr. H.J. van den Herik Universiteit Maastricht, IKAT Postbus 616, 6200 MD Maastricht Tel.: (043) 3883485. E-mail: [email protected] Dr. C. Witteveen Technische Universiteit Delft, Department Informatica, Julianalaan 132, 2628 BL Delft Tel.: (015) 2782521. E-mail: [email protected]

ADVERTISING It is possible to have your advertisement included in the BNVKI/AIABN Newsletter. For further information about pricing etc., please contact the editorial office.

Technische

CHANGE OF ADDRESS Dr. R.G.F. Winkels Universiteit van Amsterdam, Rechtsinformatica Postbus 1030, 1000 BA Amsterdam Tel.: (020) 5253485. E-mail: [email protected]

The BNVKI/AIABN newsletter is sent from Maastricht. The BNVKI/AIABN board has decided that the BNVKI/AIABN membership administration takes place at the editorial office of the Newsletter. Therefore, please send address changes to: Editorial Office BNVKI/AIABN Newsletter Universiteit Maastricht, FdAW, Vakgroep Informatica, Postbus 616, 6200 MD Maastricht, Nederland, Tel: ++31-(0)43-3883477 E-mail: [email protected]

Dr. S.-H. Nienhuys-Cheng Erasmus Universiteit Rotterdam, Informatica Postbus 1738, 3000 DR Rotterdam Tel.: (010) 4081345. E-mail: [email protected]

BNVKI newsletter

56

April 2000