The description of the network consists of two parts, a part on MLNET and a part on the CLUSTER "Computational Intelligence and Learning". MLNET is effectively a continuation of the network of excellence in Machine Learning MLnet. MLnet was active in the period 1993-1997. The CLUSTER part is a part of the work programme that concerns the cooperation between four related areas, each represented by a network of excellence: Machine Learning, represented by the actual MLNET, evolutionary computing, represented by EvoNet, neural network computing, represented by NEuroNet and fuzzy logic represented by ERUDIT. Each chapter contains a section on MLNET and a section on the CLUSTER "Computational Intelligence and Learning".

The general aim of MLNET is to support the invention, analysis, technological development and application of flexible, adaptive and knowledge intensive information technologies. These technologies encompass techniques and methods from the fields of Machine Learning, Knowledge Acquisition and Knowledge Modelling, Knowledge Discovery and Data Mining in Databases, Case-Based Reasoning and Learning, Learning Robots and Statistical Techniques for analysis of large data sets.

The main long term goals of MLNET are :

1. to stimulate and support the transfer of Machine Learning and related technologies from academia to industry,
2. to stimulate the translation from industrial needs to technical and scientific problems,
3. to represent and coordinate the wide variety of ML related research areas, which include: data mining and knowledge discovery in databases, knowledge acquisition, case based reasoning, learning robots statistical approaches to learning,
2. to disseminate information, research and application results about ML and related fields,
3. to represent the European ML community world wide,
4. to enhance the training in ML and related technologies at the European level,
5. to improve cooperation with related fields, in particular Inductive Logic Programming, Data bases, Neural networks, Evolutionary Computing, Natural Language Processing and Fuzzy Logic.

Specific goals of MLNET are:

• Maintaining and extending an information service for scientists, practitioners and students of Machine Learning. Electronically, information will be made available about scientific events and developments, courses, companies and research institutions, materials for research such as software and benchmarking problems.
• Increasing the number of industrial nodes in MLNET.
• Producing an overview of current research and development activities and an outline of main technical and scientific issues to be addressed in the near future, in the form of a Technological Roadmap.
• Demonstrating to industry how Machine Learning can be applied to practical problems in Data Mining and Adaptive Systems and identifying problems in industry that can be addressed in scientific research.
• Improving the collaboration between ML and related areas such as databases, natural language processing, case-based reasoning, knowledge acquisition, neural networks, evolutionary computing and fuzzy logic.

To achieve these goals, specific tasks will be carried out by committees. The nature and scope of these activities will be defined by a steering committee that will also monitor progress. The network will be open to institutes and companies that are active in the area of the network and that are prepared to take part in network activities.

The scientific community will benefit by a better overview of ongoing research and better access to resources such as documents, software, datasets, teaching materials. Industry will benefit by better access to available technology and increased awareness of the use of computational learning techniques.

The goal of the CLUSTER Computational Intelligence and Learning is to achieve scientific, technical and "social" integration of four communities that perform research, development and application in four related areas: machine learning, fuzzy logic, evolutionary computing and neural net computing. Networks of excellence for these areas exist and have existed for some time. Actions of the CLUSTER Computational Intelligence and Learning will be aimed at educating networks about the concepts and techniques of other networks, comparing techniques and theory, searching for possibilities to improve access by industry to a wider range of methods and tools than is provided in the context of a single network and to integrate techniques from different fields into new new techniques. In particular an inventory will be made of the needs for new technology in industry to match this with available technology at the scientific nodes. This will be achieved by:

• Organising meetings on the use of adaptive computing systems in particular sectors in industry. This should broaden the scope of methods and systems that are applied in practice and lead to a better understanding of the strengths and weaknesses of various approaches. Structuring this by sector is more effective than structuring this by scientific approach or technique.
• Improving (electronic) communications between networks.
• Providing tutoring materials for students, practitioners and scientists from other areas. This should make it possible to become acquainted with an area and provide a basis for access to the scientific literature and for the use of tools.
• Outlining research and development directions aimed at integrating techniques from the participating networks. Theoretical work is necessary for a more basic understanding of the relations between different types of techniques and the way to select one or to combine techniques.

Scientific benefits will be theoretical, technical and "cultural" integration of the fields and networks participating in the CLUSTER Computational Intelligence and Learning. Industry will benefit by gaining access to a wider range of techniques and better understanding of the applicability of tools.

The current historical moment is a delicate one for Machine Learning. In fact, there are two opposite emergent trends: on the one hand, there is a need, and a consequent request, to move ML methodologies from the academic environment to large scale applications. To satisfy this request, joint efforts to build up effective and flexible systems, by integrating methodological contributions both from inside and from outside the field, are likely to be necessary.

However, this requirement of unification is contrasted by the opposite tendency, which we are experiencing, of a fragmentation of the field into a multiplicity of rather isolated communities, each privileging a specific type of approach. This tendency has the further negative effect of favouring duplicated exploration of the same paths, with a consequent waste of useful energies.

Another relevant aspect, recently becoming apparent, is the sudden broadening of the potential applications domains for ML techniques: KDD (Knowledge Discovery in Databases), Data Mining in a number of industrial, commercial and legal domains, molecular biology, natural languages, learning from texts, information retrieval, just to mention some. The complexity of the problems requires the joint work from several disciplines, including experts from the application domain. The task of MLNET is to demonstrate the potential of its technologies for such applications and to embed them in practical methodologies and tools.

We may also notice a more concrete interest in studies on learning in Cognitive Sciences, with the possibility of joint research projects. In this variegate landscape, an institution such as a Network of Excellence will play a fundamental role in keeping connections, promoting cooperation and broadcasting information both among the emergent research themes, and between the developers and potential users of ML programs.

The mission of the CLUSTER Computational Intelligence and Learning action is to achieve closer cooperation between the areas of machine learning, case-based reasoning, knowledge acquisition (in MLNET), fuzzy logic (ERUDIT), evolutionary computing (EvoNet) and neural network computing (NEuroNet). There is an overlap in potential applications for these methods and theoretically there are relations between different paradigms that can be explored. The terminology and methodology of these areas is rather different and this makes collaboration and integration difficult. The strategy to achieve this is to explain the main concepts and methods underlying tools and techniques in the various fields to the other participants and to exploit the elements that have been created by the participating networks (teaching materials, contacts with industry and last but not least: electronic information services). This should lead to greater awareness of other techniques, to more collaborative research and to integration of tools and techniques.

The activities of MLNET will consist of a set of actions defined below. Some actions have an open character to leave room for initiatives that are proposed by members. Proposals for new actions must fit the general conditions and the constraints of the budget and must be approved by the Management Board.

An annual report will be produced that is readable and accessible for a wide audience. This will replace the quarterly newsletter of MLNET-I. It is intended to complement the more fluent information that will be available at the online information service. The annual report is therefore not directed at members of the network but at those who are interested in its activities such as potential new members, researchers and industries in adjacent fields, administrators of universities and science research councils, branch organisations industry. The annual report will be produced by the coordinating node. It will include an overview of network activities, a short financial report and an overview of the main developments in the fields covered by the network. The network will compile a distribution list in collaboration with other networks and members. The first annual report will be completed after 13 months (1 month after the first year). The second annual report will be completed after 25 months (1 month after the end of the second year).

MLnet-1 has developed a very successful information system that contains a variety of information including data sets, software, descriptions of the nodes in the network and their specialisations. The purpose of this task is to maintain this WWW infrastructure, to extend it to include new information and to improve it with the latest technology. Specific activities in this task are:

• Appointment of an information services editor responsible for actively finding, collecting, filtering and structuring information for the MLNET information services.
• Support the actions of the network with facilities for distributing information (such as announcements of meetings and other events organised by committees, reports of such events, drafts and working documents produced by committees, etc.)
• Establishment of a structured and rated software information repository to enhance reuse and technology transfer. This involves development of appropriate categorisations and testing of the software that is offered.
• On-line access to a structured and rated ML basic bibliography.
• Where possible, online demonstration of software. Active Web technologies (Java, Broadway, ActiveX) may allow setting up a try-out server where users can play with software without having to do any local installation at all.
• With a commercial publisher, publication of an ML software trial CD containing a collection of research, shareware or demo versions of commercial ML and data mining programs. This will primarily support users with low-bandwidth connections.
• Completing and updating the Research Data Base summarising the lines of research pursued by nodes in the network in particular to include research on Case-Based Reasoning, Knowledge acquisition and KDD. This can be done by building on existing knowledge describing current research and areas of expertise, for example the KA^2 database of knowledge acquisition research in Europe.
• Students who do a masters or PhD thesis will be encouraged to prepare a web page on their work, including keywords. These will be linked to the electronic information server.
• The teaching section of the information server will be extended with a page for teaching materials on ML, Knowledge Acquisition and Case-Based Reasoning.
• Support for activities of the network, such as workshops, with WWW based technology.
• Maintaining a current mailing list.

GMD will continue to provide its WWW server fee of charge. To maintain and extend the information service requires 0.5 person/year.

A document will be produced outlining the current State of the Art of the field in Europe, outlining the most promising lines of research, development and application. This can then be used as a basis for coordinating research. The Technology Roadmap will be a major deliverable of the Network. A preliminary version of a Roadmap has been produced during the preparation of MLNET-II and this can act as a starting point. After about four months a meeting will be organised (the "MLNET summit") with senior researchers in the field, who will present their views of the field. These presentations will be used by the coordinator of this action to produce a written Roadmap document. The Roadmap will include an overview of the state of the art of scientific work, an overview of applications, main research directions and main technical "needs" from an industrial viewpoint. The Roadmap will be completed before the end of the first year. After about 18 months another meeting will be held to discuss and revise the Roadmap and on the basis of this an updated version will be written. This action will be carried out apart from the corresponding action in the CLUSTER Computational Intelligence and Learning but obviously, the first version of the Technological Roadmap that is constructed there, will be used in preparing the second MLNET Roadmap. It is a significant job to do this for Machine Learning and it is even more complicated to incorporate Case-Based Reasoning and Knowledge Acquisition. Producing and integrating all these roadmaps in one pass seems a job to be tried after this network has ended and the two roadmaps are stabilised. The budget for this activity will be spent on travel to the MLNET summits and on writing the document. The budget is based on the idea that the MLNET summits can be combined with meetings of the Management Board, so that only an additional travel budget is needed.

Schedule:

Month 4: MLNET summit 1

Month 9: Roadmap 1

Month 18: MLNET summit 2

Month 21: Roadmap 2

On an ad hoc basis the network will support or elicit proposals to organise workshops that help to achieve the goals of the network. As much as possible these meetings will be held in the context of other events such as conferences to enable representation of MLNET at these events. Written proposals for such meetings can be submitted to the management board that can allocate funds for organisation or travel. A proposal must include a motivation, a detailed plan and a budget. Proposals will be evaluated by the extent to which they contribute to the goals of the network.

In cooperation with other fields that provide inductive techniques for adaptive systems (see CLUSTER part), MLNET will produce a document with guidelines for the application of Machine Learning and related technologies to practical problems. The purpose of this is to assist people from industry in evaluating if inductive techniques are suitable, and that provides methodological support for collecting and preparing data, selecting techniques and tools, applying these and interpreting the results. The document will include examples of successful projects. The budget for this action is to be used for meetings and for personnel to prepare a document (in hardcopy or electronically). The budget for this task is for organising meetings and personnel costs for writing the guide.

Month 9: Draft 1

Month 15: Final version

One of the goals is to explain and demonstrate the potential of machine learning to industry, to (events of) other sciences and to meetings on the coordination of research. This is primarily directed at audiences who are not familiar with these techniques. The Ambassador fund is to support travel with this purpose. A member of an MLNET node can apply for travel funding.

Organising education is not the main task of MLNET but coordinating and facilitating courses at a European level is part of its coordinating task. MLNET will support educational actions at a European level that complement existing education provided by universities and companies. Examples of such activities are summer schools and distance learning courses aimed at a wide European audience. A node can submit a proposal for support of such activities to the management board and this can decide to allocate financial support.

Most actions described before have the character of dissemination of information and therefore no additional dissemination actions are needed.

MLNET makes a distinction "nodes" and "main nodes". "Main nodes" are nodes that have a sustained record of good research, development or applications and that contribute significantly to actions of the network. MLNET will start with the main nodes of MLnet - 1. The set of main nodes will be extended during the period in which MLNET is active. In particular it will include main groups in Case-Based Reasoning and Knowledge Acquisition.

The management structure of MLNET consists of (a) a management board with representatives of all main nodes and that will include representatives from each subcommunity, (b) a coordinator and (c) committees that undertake specific actions. MLNET will start with a list of initial recognised subcommunities: (Knowledge acquisition and Case Based Reasoning). These subcommunities identify with the goals of the network, are of significant scientific size and stability and have already elected a representative for the MLNET management board. Subcommunities entirely within Machine Learning will not normally be recognised as subcommunities of this type, but will be represented by the main nodes as in MLNET-I. The list of recognised subcommunities will be evaluated by the management board annually. New subcommunities can organise themselves and apply for the status of recognised subcommunity which will be granted by the management board provided there are both a significant level of technical activity within the community together with a significant amount of overlap with the goals of MLNET. In this fashion, MLNET’s management structure will continue to evolve at the same pace as the scientific community evolves. By this means, smaller subcommunities will have access to the benefits of the network without having to propose their own networks, and the network will continue to include the best and most relevant efforts related to Machine Learning.

Initial main nodes: IIAI Barcelona, Daimler-Benz AG, GMD, Syllogic BV, University of Aberdeen, Universiteit van Amsterdam, Universitat Dortmund, University of Dublin, University of Ljubljana, Universite Paris-Sud and Universita di Torino.

The management board will consist of representatives of all nodes and will include representatives of each subcommunity. The initial management board will consist of:

Prof. B.J. Wielinga (Universiteit van Amsterdam, Knowledge Acquisition)

Prof. L. Saitta (Universita di Torino, A3: Technological Roadmap)

Dr. S. Wrobel (GMD, A2: Online Information Service)

Prof. M. Keane (University of Dublin, Case-Based Reasoning)

Prof. Y. Kodratoff (Universite Paris-Sud)

Prof. K. Morik (Universitat Dortmund)

Dr. R. Nakhaeizadeh (Daimler-Benz AG)

Dr. P. Adriaans (Syllogic BV)

Prof. I. Bratko (University of Ljubljana)

Drs. M. van Someren (Universiteit van Amsterdam, coordinator)

The management board will meet at the beginning of the first year, at the end of the first year and at the end of the second year of the network. In the meantime communication will be by email.

The task of the coordinator is to handle the financial administration, to coordinate reporting to the CEU, to organise the work, meetings and email contacts of the management board.

The actions of MLNET will be carried out by commitees that consist of nodes of the network. For each action a single node will be the main responsible. Beside the actions described in section 3 above, new actions can be proposed by nodes or groups of nodes. Two important tasks are defined in advance here and a budget is allocated to them: electronic communications and coordination. For other actions defined above an indicative budget is made. When an action involves financial support by the network, the node that undertakes an action presents an action plan and a budget to the management board. If this is approved, the node becomes responsible for this. The management board will meet two times a year. Meetings will be prepared and convened by the coordinator. To reduce travel costs, these meetings will be combined when possible with other events such as conferences or other activities of the network.

An institute can join the network if shows evidence of a lasting interest in the Machine Learning and related fields, contributions to these fields and the intention to contribute to the goals of the network. A group or institute that wants to join MLNET-II can apply. The application will be evaluated by the management board and on the basis of this the management board will vote. Since the management board will not meet frequently, in general this will be handled by email. Eligibility for funding will of course be subject to EEC regulations.

Nodes that do not contribute to the goals of the network and that show strongly reduced activity on the areas of the network can be removed. Once a year the management board makes a brief evaluation to check if any node satisfies these criteria. At any time a node can withdraw from the network.

The management board can invite a node to become a main node.

Some activities that are of key importance have been assigned to particular nodes:

A1: Universiteit van Amsterdam

A2: GMD

A3: Universita di Torino

The coordinator and the management board assign each of the other actions mentioned above to a node. Nodes can propose new actions of the network. Such a proposal involves a workplan, participants, results and a budget (if necessary). Substantial activities will be evaluated by the management board and minor activities by the coordinator.

Progress on an action will be monitored by two members of the management board (who do not participate in the action) and the coordinator. If not enough progress is made, this is reported to the management board that can decide to withdraw funding and assign the action to a new committee.

The cluster actions will be managed by a management board that consists of one representative of each participating network. Each action of the cluster will be assigned to a node in one of the participating networks. This node will be responsible for performing the action and producing the required result. New actions can be proposed to the management board by any node of one of the networks. Budgets for activities will be allocated as specified above. The management board will meet twice a year, when possible in the context of relevant event.

The management board of the CLUSTER Computational Intelligence and Learning actions will consist of prof. Zimmerman (ERUDIT), prof. Fogarty (EvoNet), prof. Tervor Clarkson / Dr. Mark Plumbley (NEuroNet) and Drs. M. van Someren (MLNET)

For the following actions nodes and team leaders have been appointed:

C1: Technological Roadmap: ELITE (Prof. Zimmerman)

C2: Workshops and visits: this will be managed by the management board. Proposals can be submitted to the coordinator. Every three months, the management board takes decisions about these, during a management board meeting or by email

C3: Competition: this will be managed by KCL (Mark Plumbley)

C4: Teaching materials: this will be managed by (Universiteit van Amsterdam OR Universitat Dortmund)

C5:a2

Communication: this will be distributed over the nodes in the participating networks that are in charge of (electronic) communication apart from one node (Napier Univ., prof. Fogarty) that will be responsible for the CLUSTER specific communication.

At the end of the first year and of the second year, a management report will be written. For the activities above reports and other deliverables have been specified. The project has a duration f two years. We give the actions and deliverables for the first year and the second year. The nature of these is specified in the description of the actions. We summarise this:

After each year a management report wll be produced by the coordinator. The reports on other activities are described above at the descriptions of the activities. Here we give an overview. This overview applies to both years.

The Network of Excellence will be active for 24 months. The budget of MLNET will be (figures in KECUs):

Tasks A1 will be carried out by the coordinator, Universiteit van Amsterdam. Task A2 will be carried out by GMD (Germany). Tasks A3, A4 and A7 will be delegated to nodes of MLNET.

The following nodes applied at the time the proposal for the MLNET network was sent to the CEU but did not sent their administrative data in time for the finalisation of the network programme:

Logic Programming Associates (UK)

Electricite de France (France)

Politecnico de Milano (Italy)

CSELT (Italy)

Katholieke Universiteit Brabant (Netherlands)

Vrije Universiteit Brussel (Belgium)

ISoft (France)

Imperial Cancer Research Fund (UK)

To date the nodes in ERUDIT, EvoNet and NEuroNet are those listed below.

ATLAN-tec KG

British Aerospace British Telecom Plc Cap Gemini Nederland BV CNET Daimler-Benz AG Danet GmbH Dassault Aviation

Estudio Atlas SL

EVIS Technologies GmbH

Fraunhofer Society

Genetica - Advanced Software Architectures srl

Hewlett-Packard Labs

Hollandse Signaalapparaten BV

Infoware Informatik Technologien GmbH

Intelligent Applications Ltd

National Aerospace Laboratory (NLR)

Parsytech Eastern Europe

Quadstone Ltd

Rolls Royce plc

Rover Group Ltd

Sentient Machine Research BV

SGS-Thomson Microelectronics

Siemens AG

Space Engineering SpA

Tensing GeoInformatica

Ecole Normale Superieure

Ecole Polytechnique

Napier University

Politecnico di Milano

Technische Universitaet Berlin

University of Dortmund

University of Granada (Department of Computer Science & Artificial Intelligence)

University of Granada (Department of Electronics and Computer Technology)

University of Sussex

University of the West of England

University of Vaasa

Vrije Universiteit Brussel

Aarhus University

Aristotle University of Thessaloniki

Ashikaga Inst of Technology

Christian-Albrechts University of Kiel

Coventry University

De Montfort University

Ecole Nationale de l'Aviation Civile

Ecole Polytechnique Fédérale de Lausanne (EPFL)

EERIE-EMA

Foundation for Research and Technology (FORTH)

Gesellschaft zur Foerderung angewandter Informatik (GFaI)

GMD

Hellenic Complex Systems Laboratory

INRIA

Institut d'Informatique et de Mathématiques Applicquées de Grenoble (Institut IMAG)

Institut Francais du Petrole (IFP)

Institute for Research on Parallel Information Systems - National Research Council of Italy (IRSIP-CNR)

Jozef Stefan Institute

Leiden University

National Research Council of Italy (CNR-IRSIP)

Nottingham University

Politecnico di Torino

Technical University of Szczecin

Universidad de Malaga

Universidad Pontificia Comillas

Universidade do Algarve

Universita' di Milano

Universita' di Parma

Université de Tours

Université du Littoral

Université Libre de Bruxelles

University of Goettingen

University of Birmingham

University of East Anglia

University of Edinburgh

University of Essex

University of Exeter

University of Glasgow

University of Illinois at Urbana-Champaign

University of Lasi

University of Liverpool

University of Mannheim

University of Oxford

University of Plymouth

University of Reading

University of Sheffield

University of Southampton

University of the Basque Country

University of Wales