Applied Intelligence 20, 7–8, 2004c© 2004 Kluwer Academic Publishers. Manufactured in The United States.

Guest Editorial: Machine Learning Policies

ULISES CORTES AND MARIO MARTINDepartament de Llenguatges i Sistems Informatics, Universitat Politecnica de Catalunya,

Jordi Girona 1-3, 08034 Barcelona (Catalunya), Spainmmartin@lsi.upc.es

1. Introduction

We are glad to introduce to the readers of Applied In-telligence: The International Journal of Artificial In-telligence, Neural Networks, and Complex Problem-Solving, a selection of the best papers from the 3rdCongress of the Catalan Artificial Intelligence Asso-ciation (CCIA’2000) [1]. This conference was held inVilanova i la Geltru, from the 5 to the 9 of Octoberof 2000, at the premises of the Technical Universityof Catalonia [2]. Topics covered range from theoreti-cal aspects (e.g., Machine Learning, Logics, Qualita-tive Reasoning, Fuzzy Sets Foundations, Neural Net-works) to more practical matters (related with computervision, electronic commerce, human tissue transplan-tation). This postpublication of the selected paperscontinues the trend initiated at the first Congressof the Catalan Artificial Intelligence Association(see [3, 4]).

For this special issue we had initially selected 11papers from the 44 accepted for the conference that re-ceived more than 90 original papers. We asked the au-thors to substantially extend their works to be reviewedfor publication at Applied Intelligence. The extendedversions of the initially selected papers have been re-viewed by two additional referees. This special issuecontains the 5 papers finally accepted for publication. Itis our understanding that these papers provide a limitedbut good display of some of the most recent researchin AI in Catalonia.

The common trend among the selected papers is Ma-chine Learning methods applied to learn policies ormodels of systems in order to control them.

Martin and Geffner in their paper show how gen-eralized policies for a domain can be learned from areduced set of examples generated by traditional plan-

ners on small size problems. Policies are representedas a set of priorized rules where antecedent of the rulesdescribe conditions over the arguments of the actions.This bias jointly with the expressive power of Descrip-tion Logics seems responsible of the scaling up prop-erties of the learned policies.

Gimeno and Bejar in their contribution present anearest neighbor approach to predict time series. Theframework presented explicitly takes into account pro-cess control features which is a first step for con-sidering the creation of useful learning policies. Theapproach proposed works well on partially observ-able environments, as the test task proposed in theirpaper.

Morcego’s work describes Miga, a software toolbased on the evolution of neural networks for learningsuitable models of non-linear systems. Two advantagesof this approach (not always available in non-symboliclearning) are that the system allows the inclusion ofa priori structural knowledge about the system andalso Miga allows to extract symbolic knowledge aboutthe learned model. Modelling of non-linear systems isin most cases a previous step to learn how to controlthem.

Bermejo and Cabestany’s contribution presents anew approach for reducing variability and increase ac-curacy in learning using nearest neighbor classifiersbased on prototypes. Specifically, they define an en-semble learning method, named local averaging, ap-plied over Kohonen’s LVQ algorithm, that experimen-tally has been proved successful in syntectic and realclassification problems.

Cortes et al. paper introduces the use of multi-agentsystems and electronic institutions to model the pro-curement and assignation of human tissues and organsfor transplantation. The aim of this work is to model

8 Cortes and Martın

and automate some of the complex tasks performedby a Transplant Coordination Unit (UCTx) inside aHospital. One of the focus of this work is the learn-ing of strategies to obtain the best tissue or organ for agiven recipient. The other main goal is to specify thenorms that rule these processes.

We very much appreciate and acknowledge Prof.Moonis Ali and Applied Intelligence for this oppor-tunity to make better known some of the actual workin AI in Catalonia.

The Catalan AI Association (ACIA) [5] was foundedin the summer of 1994 and has now more than 170members that are very active in AI teaching and re-search.

We acknowledge the reviewers that have providedauthors with very valuable comments and construc-

tive suggestions to enhance their papers for this specialissue.

References

1. N. Agell (Ed.), 3rd Congres Catala d’Intel.ligencia Artifi-cial, number 22 in Bulletı de l’ACIA. Asociacio Catalanad’Intel.ligencia Artificial, ACIA, 2000.

2. Technical University of Catalonia. http:// www.upc.es.3. U. Cortes and R. Lopez de Mantaras (Eds.), “2nd Catalan confer-

ence on artificial intelligence,” Computacion y Sistemas, vol. 4,no. 1, p. 3, 2000.

4. V. Torra (Ed.), “The first Catalan conference on artificial intelli-gence,” International Journal of Intelligent Systems, vol. 15, no. 3,pp. 153–154, 2000.

5. Catalan Association for Artificial Intelligence. http://www.asia.org.