Research Quality Assessment

Transcription

1 Research Quality Assessment Department of Computer Science Faculty of Science Vrije Universiteit Amsterdam produced on June 26, 2002 prof. dr J.C. van Vliet prof. dr ir M. van Steen

2 Contents Introduction 6 Research profile of the department (A) 8 A.1 General A.2 Mission A.3 Participation in research schools A.3.1 ASCI: Advanced Graduate School for Computing and Imaging A.3.2 IPA: Institute for Programming Research and Algorithmics A.3.3 OzsL: Dutch Graduate School in Logic (Onderzoekschool Logica) A.3.4 SIKS: School for Information and Knowledge Systems A.4 Facilities A.5 Financial flexibility permitting a new research policy A.6 Human resource policy A.7 Internal quality management Key data of the department (B) 14 Description per program (C) 17 C.8 Artificial Intelligence C.8.1 Full title C.8.2 Subprograms

3 2 C.8.3 Program members C.8.4 Keywords C.8.5 Research input of academic staff (wp) C.8.6 Research output C.8.7 Composition of research input academic staff C.8.8 Program description in brief Agent Systems Computational Intelligence Intelligent Interactive Distributed Systems Knowledge Representation and Reasoning C.8.9 Overview of scientific results Agent Systems Computational Intelligence Intelligent Interactive Distributed Systems Knowledge Representation and Reasoning C.8.10 Development of the program Agent Systems Knowledge Representation and Reasoning Computational Intelligence and Intelligent Interactive Distributed Systems. 27 C.8.11 Societal/technological relevance Agent Systems Computational Intelligence Intelligent Interactive Distributed Systems Knowledge Representation and Reasoning C.8.12 Other indicators of quality C.8.13 Five key publications C.9 Information Management and Software Engineering C.9.1 Full title C.9.2 Subprograms C.9.3 Program members C.9.4 Keywords C.9.5 Research input of academic staff (wp) C.9.6 Research output C.9.7 Composition of research input academic staff

4 3 C.9.8 Program description in brief Mission statement Software Engineering Information Systems Business Informatics C.9.9 Overview of scientific results Software Engineering Information Systems Business Informatics (BI) C.9.10 Development of the program Software Engineering Information Systems Business Informatics C.9.11 Societal/technological relevance C.9.12 Other indicators of quality C.9.13 Five key publications C.10 Computer Systems C.10.1 Full title C.10.2 Subprograms C.10.3 Program members C.10.4 Keywords C.10.5 Research input of academic staff (wp) C.10.6 Research output C.10.7 Composition of research input academic staff C.10.8 Program description in brief Distributed Systems Parallel Computing C.10.9 Overview of scientific results Distributed Systems Parallel Computing C Development of the program Distributed Systems Parallel Computing C Societal/technological relevance Distributed Systems Parallel Computing C Other indicators of quality

5 4 C Five key publications C.11 Theoretical Computer Science C.11.1 Full title C.11.2 Subprograms C.11.3 Program members C.11.4 Keywords C.11.5 Research input of academic staff (wp) C.11.6 Research output C.11.7 Composition of research input academic staff C.11.8 Program description in brief C.11.9 Overview of scientific results Term rewriting Systems Software Verification Coalgebraic Techniques C Development of the program Term rewriting Systems Software Verification Coalgebraic Techniques C Societal/technological relevance C Other indicators of quality C Five key publications Publications per program (C) 65 C.12 Artificial Intelligence C.12.1 Dissertations C.12.2 Scientific Publictions Journals (Refereed) International Conferences (Refereed) Contributions to Books Books Other C.13 Information Management and Software Engineering C.13.1 Dissertations

6 5 C.13.2 Scientific Publictions Journals (Refereed) International Conferences (Refereed) Contributions to Books Books Other C.14 Computer Systems C.14.1 Dissertations C.14.2 Scientific Publictions Journals (Refereed) International Conferences (Refereed) Contributions to Books Books Other C.15 Theoretical Computer Science C.15.1 Dissertations C.15.2 Scientific Publictions Journals (Refereed) International Conferences (Refereed) Contributions to Books Books Other

7 SECTION Introduction 6

8 7 This report was prepared as input for the assessment of the quality of research in Computer Science in the Netherlands by the Association of Universities in the Netherlands (VSNU). It contains the requested documentation for the research programmes of the Department of Computer Science of the Vrije Universiteit in the period The general VSNU protocol 1998 and the specific protocol for computer science provided by the VSNU were followed as closely as possible. The Wiskundig Seminarium of the Vrije Universiteit was founded in 1930, and grew out to the Faculty of Mathematics and Computer Science in the 1970s, which still existed in Since the previous assessment the organisational embedding of computer science at the Vrije Universiteit has changed twice. In 1998 the Faculty of Mathematics and Computer Science formed the Faculty of Sciences, together with the Faculties of Physics and Astronomy, and Chemistry, incorporating the former three faculties as divisions. In 2001 it was decided to reorganize the Faculty of Sciences into four organisational units, one of which is the Department of Computer Science. Even though this reorganization will take effect only in 2002, for this report the Department of Computer Science has been chosen as the most natural unit for the assessment. The organisational reorganizations were focused on creating stronger and more professional administrative services as well as better opportunities for multidisciplinary research and education. Even though, as we shall show, the present research programme in computer science incorporates a wider area of multidisciplinary research than before, this change can probably not be ascribed to the organizational changes. The new administrative structure has had only a minor influence on the research programme in computer science. Most notably, there is a close collaboration with the Division of Physics and Astronomy, where Bal was appointed part-time professor and head of the Physics Applied Computer Science group in The joint expertise in visualization (at physics) and cluster computing is used to study interactive applications. The specific VSNU protocol for computer science gives clear rules for calculating the research input of full professors, UHDs, UDs, postdocs and aios in a standard manner. These have been followed in this report. This report was prepared by M. van Steen and J.C. van Vliet, in consultation with the new chairman of the department of computer science (J.W. Klop) and one of the members of the present board of the Division of Mathematics and Computer Science (H. Bal). The report is based on input from the heads of the four subdepartments in computer science. The report has been approved by the Board of the Faculty of Sciences, the director of the Division of Mathematics and Computer Science, and the Scientific Research Committee of the division.

9 SECTION A Research profile of the department 8

10 9 A.1 General During the period for this assessment the faculty of the Department of Computer Science has changed considerably. The most noteworthy changes have been the forming of new groups in Parallel Programming (headed by H.E. Bal), Business Informatics (headed by J.H.M. Akkermans), Intelligent Interactive Distributed Systems (headed by F.M. Brazier), and Computational Intelligence (headed by A.E. Eiben). Furthermore, the professor of Information Systems retired (R.P. van de Riet), and his successor was appointed (C. Verhoef). The formal decision to establish a new group in Bioinformatics was made in 2000; this group will become operational as of 1 September Because several of these new directions were already in preparation during the period preceding that of the present evaluation, there is a mixture between continuation in research directions, as well as the initiation of new ones. A.2 Mission The mission of the Department of Computer Science is partly described in a policy plan 1 prepared in 1999 by the four heads of the subdepartments. The strategy was put into a budget planning for the next ten years by the director of the Division of Mathematics and Computer Science (then: J. Oosterhoff). Most parts of this planning have already been implemented. The research of the Department of Computer Science is aimed at better understanding the foundations, mechanisms, and technologies that are needed to develop and deploy information and computing systems that are characterized by their complexity. This complexity is often caused by the size and distribution of data, processes, and management, such as in the case of Internet-based systems, Grid computing, or data warehouses. In other cases, complexity is caused by rich, heterogeneous, and evolving functionality that needs to be captured in a single framework, as is the case with many modern information systems. This research is largely inspired by problems that have their immediate origin in modern society. Many projects are carried out in cooperation with businesses and organizations outside our university. Much of our research concentrates on Internet-based information and computing systems, and involves questions related to multi-agent systems, Semantic Web, e-business, Grid computing, and scalable middleware. Another field of research activities, and which partly overlaps the previous one, is that of large-scale information systems in which we concentrate on methods to improve the development and deployment of such systems, as in re-engineering, software architectures, human-computer interaction and scalable verification methods. In all cases, the inherent complexity of the problems investigated plays a key role and is a source of inspiration. Many of our research activites are characterized by their multidisciplinary approach. Projects often involve collaboration with fellow researchers from different fields. This is most notably true for research related to application domains such as e-business or bio-informatics, but also holds for projects in human-computer interaction and agent technology (psychology), software asset management (economy) and scientific visualization (physics). The research activities are concentrated within four subdepartments: 1 Informatie en Communicatie via Wereldwijde Netwerken, 1999, by J.C. van Vliet, A.S. Tanenbaum, J. Treur, and J.W. Klop.

11 10 Artificial Intelligence (AI) Information Management and Software Engineering (IMSE) Computer Systems (CS) Theoretical Computer Science (TI 2 ) These four subdepartments are considered of equal importance (even though the sizes may differ), reflecting the needs of the educational curricula in Computer Science, Artificial Intelligence, and Information Science. A.3 Participation in research schools Almost all research activities are carried out as part of Dutch research schools. Only the newly appointed staff members in Information Systems are not yet a member of a research school. At present, our department participates in four different schools: ASCI, IPA, OzsL, and SIKS. A.3.1 ASCI: Advanced Graduate School for Computing and Imaging Staff member Subdept. Role in the research school Bal, H.E. CS Member of the board Grune, D. CS Member of the education committee Jonge, W. de IMSE Member Kielmann, T. CS Member Steen, M.R. van CS Member of the scientific committee Tanenbaum, A.S. CS Scientific director A.3.2 IPA: Institute for Programming Research and Algorithmics Staff member Subdept. Role in the research school Klop, J.W. TI Member of the board Raamsdonk, F. van TI Member of the education committee Fokkink, W. TI Member of the scientific committee Rutten, J.M.M.M TI Member A.3.3 OzsL: Dutch Graduate School in Logic (Onderzoekschool Logica) Staff member Subdept. Role in the research school Vrijer, R. de TI Member Rutten, J.M.M.M TI Member 2 From the Dutch Theoretische Infromatica.

12 11 A.3.4 SIKS: School for Information and Knowledge Systems Staff member Subdept. Role in the research school Akkermans, J.M. IMSE Member of the board Brazier, F. AI Member Bruin, H. de IMSE Member Gordijn, J. IMSE Member Top, J.L. IMSE Member Dormann, C. IMSE Member Hulstijn, J. AI Member Eiben, A.E. AI Member Eliëns, A IMSE Member Fensel, D. IMSE Member Harmelen, F. van AI Member Jonge, W. de IMSE Member Jonker, C. M. AI Member Kowalczyk, W. AI Member Lämmel, R. IMSE Member Marchiori, E. AI Member Stebletsova, V. AI Member Riet, R.P. van der IMSE Chair of the board of governors Treur, J. AI Member Veer, G.C. van der IMSE Member Vliet, J.C. van IMSE Chair of the scientific committee A.4 Facilities The department provides an excellent research environment, with a positive atmosphere. Travel of its research staff is stimulated and sufficiently funded. Opportunities for inviting guests from abroad are ample and are important to foster international collaboration. The computer network and computational support have traditionally been organized in cooperation with the Department of Mathematics, and is excellent. All faculty members and graduate students are provided with both a desktop and a laptop computer. The situation of the library at the Vrije Universiteit has not significantly changed since the previous review, and it is still small in size. The funding of the library, which is not under the direct control of the Department of Computer Science, is unlikely to increase significantly in the coming years. Fortunately, the most important journals and books are available, online access is provided to the digital libraries of the ACM, the IEEE Computer Society, and the computer science journals of Elseviers. Other journals or books can be requested from other libraries, and the library of CWI is an excellent alternative, albeit at a travelling distance of up to one hour.

13 12 A.5 Financial flexibility permitting a new research policy Over the past years, the department has been able to realize a moderate growth in its budget for direct funding ( eerste geldstroom ). This growth is largely due to an increase in the total number of students enrolled, which in turn is the result of various initiatives to start new, or expand existing, curricula (for example in the area of Information Science). This capacity has been used to set up various new groups, such as those in Business Informatics and Computational Intelligence. The budget for direct funding is largely allocated to permanent staff positions. Turnover in permanent staff is low, and these rare opportunities are used to carefully consider our research policy, and redirect research whenever this is deemed possible and worthwhile. The department has one permanent postdoc position and thirteen PhD positions. Furthermore, the department has succeeded in acquiring a large number of additional grants from NWO and thorugh contract research. The department s own non-permanent positions as well as those from grants are used to realize a regular influx of new people and new ideas. Three substantial grants deserve to be mentioned explicitly. Bal received a PIONIER grant ( ), and during , his research is funded by a USF (Universitair Stimulerings Fonds) grant of the VU. These grants allowed us to significantly increase the research staff of the group on Parallel Computing. A substantial grant from Stichting NLnet made it possible to start the new group in Intelligent Interactive Distributed Systems. A.6 Human resource policy Notwithstanding the small numbers of qualified candidates the department has managed to attract good students to all open positions for PhD students. The master programmes for foreign students play an increasingly important role in attracting these students. Besides the standard salary during 4 years, every PhD student is awarded 5000 euros for travel and books during the course of the studies. A research plan and agreement is written at the beginning of the PhD term. PhD students receive training directly from the supervising professor, and in many cases are required to actively participate in small groups of students and staff members working in the same subject area. Furthermore, PhD students take part in the educational programmes of the research schools or other networks, and are likewise generally offered to participate in a relatively intensive course on writing scientific texts. We have also succeeded in allocating the available postdoc positions to excellent candidates, although sometimes requiring considerable effort. Postdoc researchers actively participate in research programmes, and are often closely coupled to the work that is being carried out by one or more PhD students. In this way, they are trained to carry out supervision. In most cases, postdoc research is carried out under supervision of, and in close collaboration with the professor responsible for the research programme. A.7 Internal quality management The chairman of the scientific committee is charged with monitoring the progress of all PhD students in computer science. This is formalized through written progress reports after 6 months, and 1, 2 and 3 years, and conversations both with the student and his/her advisor after 6 months and 1 year. (In the last two years the changes in the administrative structure of the department have led to less adherence to this system, but the system is revitalized in 2002.) If necessary an appointment as PhD student can

14 13 be discontinued after one year. The research schools also play a role in monitoring the progress of PhD students. The four heads of the subdepartments conduct yearly, formalized functioneringsgesprekken with all members of their subdepartment. The research activities and directions are among the most important subjects of these reviews. For the full professors similar evaluations are conducted by the dean of the Faculty of Sciences.

15 SECTION B Key data of the department 14

16 15 1a University: Vrije Universiteit 1b Faculty: Faculty of Sciences 1c Section/Subfaculty: 2 Departments: Computer Science 4a Research institutes: (commissioner) 4b Research institutes: (participant) 4a Research schools: SIKS: School for Information and Knowledge Systems (commissioner) 4b Research schools: ASCI: Advanced Graduate School for Computing and Imaging (participant) IPA: Institute for Programming Research and Algorithmics OzsL: Dutch Graduate School in Logic (Onderzoekschool Logica) SIKS: School for Information and Knowledge Systems 5 Education programmes: Artificial intelligence (Kunstmatige intelligentie) Computer science (Informatica) Information science (Informatiekunde) 6a Income in 2001: Direct funding kf NWO funding kf 700 Contract research kf b Costs in 2001: Personnel kf Other kf a Human resources 2001: Academic staff Supporting staff b Students in 2000/2001: 580

17 16 8 Research programmes: 1 Artificial Intelligence 48.6 fte 2 Computer Systems 50.0 fte 3 Information Management and Software Engineering 58.2 fte 4 Formal Methods and Software Verification 23.8 fte Total: fte

18 SECTION C Description per program 17

19 Program: Vrije Universiteit / Artificial Intelligence Research director: prof. dr J. Treur 18 C.8 Artificial Intelligence C.8.1 C.8.2 C.8.3 C.8.4 C.8.5 Full title Subprograms Program members Keywords Research input of academic staff (wp) FTE Total wp1 phd other wp2 phd other wp3 phd other Total C.8.6 Research output Type Total PhD Theses 3 3 Journals Conferences Book chapters Books 1 1 Other Total PhD Theses: Number of completed Ph.D. theses Journals: Number of publications in refereed journals Conferences: Number of publications at refereed international conferences Collections: Number of refereed contributions to scientific books Books: Number of scientific (text)books Other: Number of other types of scientific publications C.8.7 Composition of research input academic staff 2001 C.8.8 Program description in brief The Artificial Intelligence research programme aims at an integration of fundamental and applied research. Various real-world application domains offer problems that require adequate techniques to be

20 Program: Vrije Universiteit / Artificial Intelligence Research director: prof. dr J. Treur 19 addressed. These requirements introduce problems to be investigated in fundamental research. Techniques available from fundamental research usually provide partial solutions; further development in fundamental research is needed to obtain more adequate solutions. The programme addresses four subprogrammes, more in particular: Agent Systems, Computational Intelligence, Intelligent Interactive Distributed Systems, Knowledge Representation and Reasoning. Agent Systems Within the Agent Systems subprogramme it is investigated which supporting methods and techniques are adequate for the development of intelligent agent systems. Intelligent agents are often meant to be used within information- and knowledge-intensive domains and are required to show nontrivial types of behaviour. To design such agents standard methods from Software Engineering are insufficient. More dedicated methods and techniques are required to deal with the behavioural complexity of intelligent agents and to incorporate knowledge modelling techniques. More specifically this subprogramme has as objective conceptual modelling methods and techniques to support analysis (requirements specification, verification) and design (including prototype simulation) of agent systems, reusable models for a number of application domains, and formalisation and foundation of such techniques and models. Examples of application domains addressed are brokering, negotiation, information agents, design agents, project coordination and multi-agent organisation. For such application domains real-world case studies are undertaken. By iterative development and testing of languages, models, templates/library and techniques, further progress is made. Expected results aimed at included: a specification language for component-based design of (intelligent, knowledge-intensive) agent systems a software environment for design and prototype simulation a behavioural requirements specification language for agent systems a compositional verification method for agent systems a software tool to (model) check traces against specified behavioural requirements formalisation and foundation reusable templates for agent design models, coordination and negotiation models, behavioural requirements and proof templates, and libraries of such templates Computational Intelligence The subprogramme and the corresponding group Computational Intelligence (CI) has been established in The subprogramme s objectives are the investigation of CI techniques and their application. CI as a field is very diverse with various technical approaches to realizing intelligence. The common feature within CI subareas is the non-symbolic, emergent attitude, where desired structure, solutions to problems and/or targeted system behavior emerges from bottom-up. This complements the traditional symbolic AI approach. The group s strength lies in evolutionary computing and machine learning, respectively, data mining. Computer simulated evolution can be envisioned at least from two perspectives. It can be seen as a

21 Program: Vrije Universiteit / Artificial Intelligence Research director: prof. dr J. Treur 20 problem solving metaphor and as a simulation framework to study complex adaptive systems behavior. Most of the research in this subprogramme is in line with the first stream ( hard core evolutionary computing), with growing interest and gradually more activities in artificial life and evolutionary economy that belong to the second type. Machine learning and data mining is a field with an intrinsic dichotomy: being a field of a certain type of problems and a collection of appropriate problem solving techniques (algorithms) at the same time. The group is active in studying new algorithms as well as investigating new application areas. Intelligent Interactive Distributed Systems The Internet is today s main infrastructure for communication between systems worldwide. Information is inherently distributed over such systems on the same world-wide scale. The systems themselves may be positioned at one location or mobile; e.g., laptops, and intelligent PDAs. Software agents provide a means to access information and/or services that are inherently distributed, but not necessarily accessible from all parts of the world due to, for example, bandwidth constraints, processing constraints and/or security constraints. Agents are, in fact, multi-threaded processes that autonomously roam the internet, gather and process information, and interact with other agents and objects. Agents may, or may not, have an external, public embodiment: public information that can be accessed by others. Their processes are not visible to the outside world. As the number of agents and objects in a multi-agent system increase, scalability becomes of ever increasing importance. Multi-agent systems need to be able to scale (in terms of agents and available resources) on demand. This needs to occur without noticeable loss of performance or considerable increase in administrative complexity. They also need to be robust, capable of dealing with the dynamic nature of the Internet, securely. The Intelligent Interactive Distributed Systems Group (IIDS Group) was initiated by Stichting NLnet and established by the Vrije Universiteit January 1st, The group s mandate is to advance technology to support the development of flexible, adaptable architectures for large-scale intelligent interactive distributed systems: technology needed to cope in a world in which the Internet plays such a significant role. The current focus is the AgentScape framework (this research is done in collaboration with the Computer Systems Group at the Vrije Universiteit) and the legal implications of agent technology. Knowledge Representation and Reasoning The subprogramme and corresponding group Knowledge Representation and Reasoning (KR&R) has been growing as a separate subgroup with the AI research programme since Knowledge Representation is the foundational area within Artificial Intelligence that studies fundamental and theoretical properties of methods for symbolically representing and manipulating knowledge. This field is firmly placed in the symbolic branch of AI, and typically uses formal logic (in the broad sense of the word logic ) as its guiding paradigm. Besides logic for representation, the second major ingredient of KR&R is the use of computation for manipulating the logical symbols to derive new information. In this sense, KR&R is essentially different from logic, since it is not only concerned with which inferences are sanctioned, but also with which inferences can be efficiently made. Besides foundational work, the KR&R group at the VU has strong interests in applications

22 Program: Vrije Universiteit / Artificial Intelligence Research director: prof. dr J. Treur 21 (Knowledge-Based Systems, Semantic Web) and methodology (Knowledge Engineering). The KR&R group has a very international orientation. Besides national funding from NWO, most funding is acquired from EU projects. In the period the group has been involved in the following projects: approximate notions of programme correctness (NWO OIO), safety-critical knowledge-based systems (NWO postdoc) On-To-Knowledge: Content-driven knowledge management through evolving ontologies (EU/ISTproject) IBROW: Intelligent Brokering Service for Knowledge-Component Reuse on the World Wide Web (EU/FET-O project) Protocure: Improving medical protocols by formal methods (EU/FET-O project) The OntoWeb Network (EU/IST thematic network) More on these and other (current) projects can be found at ai/projects/knowledge-representation.html. In the period the group has employed staff from Spain, Maroco, Germany, and has hosted visitors from UK, France, Austria. In this period, papers have been co-authored with researchers from Manchester, New York, Stanford, Karlsruhe, Bremen and others, as well as in the Netherlands with researchers from Groningen, Erasmus, UvA, Utrecht. C.8.9 Overview of scientific results Agent Systems Within the period , the Agent Systems subprogramme has led to 5 Ph.D. theses, whereas for autumn 2002 another 3 are planned. The following results were achieved. Design of intelligent agent systems A component-based design method (DESIRE) has been developed for design of (knowledge-intensive) agent systems. This method includes a language for design specification at a conceptual modelling level and software support has been developed for design and automated prototype simulation. Methodological and semantic principles underlying the design method and component-based architecture have been identified and worked out in detail. A number of publications cover aspects of this design method, as do the Ph.D. Theses of Niek Wijngaards (1999) and Pascal van Eck (2001). For further support a library icluding a large number of reusable generic component-based design models has been developed and applied for different types of agents: Generic Agent Model GAM, applied to model and simulate animal behaviour, and a society of simple agents; furthermore it was used as a basis for refinement for many of the other, more specialised agent models. cognitive agent models: BDI-agent model, a normative agent, a model for trust dynamics, a model for mind-matter interaction (with applications in cooperation with cell biologists) and a

23 Program: Vrije Universiteit / Artificial Intelligence Research director: prof. dr J. Treur 22 model interaction between verbal and nonverbal communication between agents; Ph.D. Thesis of Wouter Wijngaards (2002). models for social agents functioning within a multi-agent system: a broker agent model, applied in Electronic Commerce applications in projects ABMP and ICEBERG both funded by Dutch Telecom industry KPN, and in projects in cooperation with insurance industry and IT industry Ordina Utopics and CMG; a cooperative agent model, applied in a SENTER project co-funded by Rabobank addressing work flow management and distributed agenda scheduling in relation to a Call Center; a multi-attribute negotiation agent model, applied in the Electronic Commerce project ABMP funded by KPN. a self-maintenance agent model and an explanation agent model, applied in a project funded by DSM chemical industry. Moreover, a number of component-based task models have been developed, including models for diagnosis (applied in diagnosis of Nylon production in a project funded by DSM), design (Ph.D. Thesis of Pieter van Langen (2002)), and process control. We have also developed a generic deliberative evolution model for dynamic (re)design and (re)configuration of agent systems (at run-time), and incorporated this model in the DESIRE software environment (see Ph.D. thesis Niek Wijngaards, 1999). Based on the underlying concepts of DESIRE, in a project funded by ATO, a simulation environment has been developed for configuration and execution of mathematical quality (of perishable agricultural products such as vegetables) simulation models from model components (see, e.g., Ph.D. thesis of Mark Sloof, 1999). Before the Computational Intelligence subprogramme was started, within the Agent Systems subprogramme some interesting results in Datamining were achieved. A new technique for analyzing data, Rough Data Models, was developed that was successfully applied for solving numerous data mining problems, like fraud detection, retention modeling, mailing selection, application credit scoring. This technique is currently used by companies like the Robeco Group, Interpay, Visa Card Services. Additionally, Rough Data Models, combined with techniques of digital signal processing, led to a system for automatic interpretation of animal vocalizations (cow moo s). Moreover, a heuristic algorithm for solving constraint satisfaction problems was developed. Analysis: requirements specification, simulation, checking and verification For analysis of agent models and task models the expressive behavioural requirements specification language TTL (Temporal Trace Language) has been developed, that, e.g., allows for specification of relative adaptive behaviour in which comparison between two different histories has to be expressed. To manage complexity the language allows for abstraction, by means of abstract language constructs that indicate complex dynamic patterns in a user-defined manner; e.g., Ph.D. Thesis of Wieke de Vries (2002). A software tool to model check behavioural traces against specified requirements has been developed. A number of templates for dynamic properties and their logical relations (proof-templates) have been identified and collected. An executable sublanguage of TTL has been developed that provides an adequate language for declarative specification of simulation models. A simulation environment has been developed to support execution of these simulation models; e.g., Ph.D. Thesis of Wouter Wijngaards (2002). Furthermore, a compositional verification method for agent systems has been developed that allows to relate dynamic properties of an agent system as a whole to properties of the agents, and/or properties of a component-based agent model or complex task model to properties of its components. This method has been used to identify dynamic properties of cooperating agents, and relate

24 Program: Vrije Universiteit / Artificial Intelligence Research director: prof. dr J. Treur 23 them to succesfulness properties of cooperation for an agent system as a whole. Moreover, the basic ideas behind the compositional verification method have been exploited to identify properties within a multi-agent organisation model at different levels of aggregation, and to relate these properties from different levels to each other. In addition, as an extension of TTL a requirement specification language has been developed to specify dynamic properties of (runtime) multi-agent system configuration dynamics. Finally, dynamic properties of beliefs, desires and intentions and their relations have been identified. Formalisation and foundation Substantial work has been done to provide a solid foundation for the basic concepts and techniques developed in the research subprogramme as a whole. On the one hand this has been achieved by providing formal definitions of the specification languages for requirements and design structures, and by providing formal specifications of requirements, design and proof templates. On the other hand semantic foundations have been defined. A first example is the semantic foundation of the component-based design models both for complex (sequential) component-based task models and for distributed agent systems (e.g., Ph.D. thesis of Pascal van Eck, 2001). A second example is the foundation of the compositional verification method. It has been shown how this method can be formalised within Temporal Multi-Epistemic Logic (e.g., Ph.D. thesis of Joeri Engelfriet, 1999). Furthermore the semantics of the dynamics of nonmonotonic reasoning processes has been addressed extensively. Some of the concepts developed, turned out useful to provide a foundation for the analysis of multi-interpretable ecological monitoring information, developed within the project EKS, funded by the Dutch Ministry for Agriculture and Nature, Natuurmonumenten, Staatsbosbeheer and IPTS. Computational Intelligence Results of this subprogramme within the period can be reviewed as follows. Evolution as problem solver Research into evolutionary problem solving involves a better understanding of various features of evolutionary algorithms (EA) and utilizing the power of EAs to tackle challenging problems. As for the first aspect investigations concern on multi-parent crossover and parameter control, in particular self-calibrating EAs adapting themselves on-the-fly to a given problem. As for the second aspect, NP hard/complete problems have been treated, such as randomly generated binary constraint satisfaction problems or satisfiability. The results have been published in various journals, conference proceedings and edited monographs. Especially interesting is the self-calibrating power of EAs (a related paper has won the IEEE paper award in 2001) and the ability to handle constrained problems where the original problem formulation supplies no optimization objectives hence preventing a straightforward evolutionary approach that requires a fitness measure. Soft evolution Evolutionary processes run in an interactive fashion based on subjective selection form a special type of applications often seen in design and art. The Mondriaan Evolver (on-line on the Web from the groups home page) generates images in the style of Piet Mondriaan serving as a show case for this kind of applications. The Escher Evolver has been implemented in-vivo during the exhibition M.C. Escher: real and Virtual in the Gemeentemuseum The Hague, May through November Evolution driven simulations in an artificial life, respectively evolutionary economy

25 Program: Vrije Universiteit / Artificial Intelligence Research director: prof. dr J. Treur 24 context have provided insights into the complex dynamic system behavior emerging from locally acting forces. Distributed evolutionary computing Research within the Distributed Resource Evolutionary Algorithm Machine (DREAM) project aims at utilizing existing but not fully used computing power via the Internet in a peer-to-peer fashion. Evolutionary computation is a very natural meta-application as evolution is intrinsically parallel and unsynchronized, forced on traditional sequential machines only by technical limitations. The peer-to-peer paradigm allows an enormous scale-up of the problem solving power. The work performed so far has laid a solid infrastructure for this type of computing and proof of principle applications. Machine learning and data mining Research in machine learning is mainly concerned with the design and application of novel learning algorithms integrating different approaches. In particular, the NWO funded project Genetic Relational Search for Inductive Learning (GeneRe) deals with the development of a system integrating evolutionary computation and inductive logic programming techniques for learning in first order logic. The work so far performed has shown the effectiveness of this integration when applied to standard benchmark problems. Moreover, a recent interdisciplinary project funded by the VU started in September 2001, that deals with the use of statistics and machine learning for analyzing data from biological experiments (bio-informatics). Application oriented research in the field of data mining addressed several practical problems: development of a fraud detection system for transactions with credit cards (Interpay), mailing selection (first place in the Benelearn 99 Competition), predicting peak electric load of a power plant (EUNITE). Additionally, we did some research on applications of neural networks to text mining (automatic classification of poems written by different poets). Intelligent Interactive Distributed Systems The group started in 2000 with 0.4 full professor. In June 2002 it now has 2 IIDS PhD students, 2 PhD students together with Computer Systems, 2 assistant professors, 1 adjunct researcher, 1 programmer and 0.8 full professor. AgentScape framework To support the development of agents the AgentScape framework (Wijngaards, Overeinder, Steen and Brazier, 2002) will include the AgentScape Operating System (AOS), directory services, an agent factory, and libraries for developers of intelligent multi-agent systems. This environment is being designed to support heterogeneity, interoperability with legacy systems, and multiple qualities of services. Agent Operating System The Agentscape Operating System is middleware designed specifically to support agents in large-scale distributed environments. It can, in fact, be seen to be a virtual machine distributed over a wide-area network consisting of heterogeneous hosts. A first prototype has been implemented and tested. Further development is on way. The Agent Factory The Agent Factory is an AgentScape service, based on the work done by Niek Wijngaards PhD thesis (see section on Agent Technology). It (re-)designs agents configurations and knowledge during an agent s lifetime. An agent can go to an Agent Factory and ask

26 Program: Vrije Universiteit / Artificial Intelligence Research director: prof. dr J. Treur 25 for its knowledge and/or functionality to be modified (e.g. updates of ontologies). The Agent Factory can also be used to extend the basic mobility of agents to true heterogeneous environments, i.e., different operating systems, different agent platforms, and different programming languages. To this purpose the Agent Factory automatically adapts mobile code to a specific host: a form of generative mobility. This work is the basis for a number of papers in the field of AI and Design, Adaptive Agents. Programming environments The WWW has limited functionality. It was designed for distributed document access, not for distributed processing. The MANSION paradigm and environment are the topic of a PhD project (Guido van t Noordende) in collaboration with prof dr A.S. Tanenbaum to support application designers using Agentscape to create multi-agent applications. Analysing Legal Implications and Agent Information Systems (ALIAS) The ALIAS project is a collaborative project with two legal departments in the Netherlands: the Computer Law Institute at the Vrije Universiteit, and the Centre for Law, Public Administration and Informatization. This exploratory project aims to identify the legal and technical implications of the use of software agents. One of the side effects of this project is closer contact with the legal field (e.g. invited speaker at a recent legal international conference). Another is input for the design of the Agentscape framework and real-life applications. Knowledge Representation and Reasoning Within this subprogramme, work in the following areas has lead to some 35 publications in international conferences and journals, and to 3 Ph.D. theses (co-supervised at either VU, UvA and Nice). Approximate reasoning Classical logic-based reasoning provides either a correct answer or no answer at all. We have studied alternative forms of reasoning which provide approximate answers under suboptimal circumstances. This allows for more gradual correctness criteria, and can be exploited in anytime algorithms. In particular in the area of approximate diagnostic reasoning we have obtained promising theoretical results. Verification of knowledge-based systems The past decade has resulted in formal specification languages for conceptual models of knowledge-based systems. These languages give a formal basis on which to perform validation and verification of KBS. Verification of medical protocols Medical protocols describe a medical procedure for a given category of patients. We are using KR languages for formally representing such protocols, and to perform various forms of reasoning with such protocols (verification, critiquing, configuration etc). Knowledge Representation for the Semantic Web The growth of the Web has resulted in a very large on-line body of weakly structured information. KR&R techniques help in searching, navigating, visualising and maintaining such weakly structured information. We coordinate a large and highly visible EU-funded project in this area, called On-To-Knowledge.

27 Program: Vrije Universiteit / Artificial Intelligence Research director: prof. dr J. Treur 26 C.8.10 Development of the program Agent Systems The Agent Systems subprogramme was developed in continuation of the programme on compositional knowledge system design in the previous assessment period. This continuation has preserved all achievements of the previous period in the sense that agents can be designed that include as a component a compositional knowledge-based system. Thus intelligent agents can be realised that are needed to address information- and knowledge-intensive domains. The emphasis in the period has shifted to the agent aspects. In the first half of the assessment period the design of intelligent agent systems was addressed. The dominating application area has been brokering and negotiation in Electronic Commerce contexts. A number of component-based proof-of-concept prototype applications have been developed as a test for the agent design method. In the second half of the assessment period the analysis (requirements specification, checking, verification) of agent system behaviour became more and more the focus. In the last years of the period, application areas were gradually moving to academic disciplines: Biology, Cognitive Science, Social Sciences and Organisation Theory, and cross-connections between these domains. As a unifying challenge within these disciplines the agent-oriented analysis and modelling of dynamics of complex but (self)organised phenomena at an abstract, conceptual level was recognized. This challenge has only been addressed partially in the last one or two years. Some of the preliminary results are: the analysis and simulation of the intracellular dynamics underlying bacterial behaviour using a temporalised agent model based on beliefs, desires and intentions (BDI), the analysis of dynamic properties of an organisation model in relation to dynamic properties of its parts, and the analysis and simulation of intracellular dynamics based on organisation modelling techniques. The challenge of dynamic modelling in these disciplines is expected to be one of the main aims of the Agent Systems research programme for the next years. Knowledge Representation and Reasoning Steady growth: The KR&R group has started out as a single permanent member of staff, a temporary lecturer and a Ph.D. student in 1996, and has recently grown to a professorship, a permanent lecturer, two post-docs and three Ph.D. students. Signficant funding is obtained from EU sources, and has been secured for the next few years. Some funding is obtained from national sources (NWO), with other sources (ICES/KIS, Token2000) being pursued. The Semantic Web has been a major growth area for the group, both in terms of funding, in term of international visibility and status, and in terms of research results. We are expecting this to remain the case in the next few years. Important research challenges will be the move from the current rather strict, formal and brittle methods to more robust methods that exploit approximation techniques and anytime algorithms. We expect to be able to build here on results that we have obtained earlier in the areas of diagnosis and planning.