Evolution and Neurons in ICT and Management
by Han La Poutré
Evolutionary methods and neural networks are important computation methods inspired by essential principles in Nature, like survival of the fittest, evolution, learning processes, interaction, and parallelism. They provide important new possibilities for a fast and good solution of problems related to economy and management. R&D at CWI focuses on evolutionary algorithms, as well as on dynamic algorithms and neural networks. The main applications are related to management and ICT, including E-commerce, on-line process management, remote sensing classification, and economy.
Evolutionary algorithms are traditionally applied to areas like mathematical optimization and automatic programming. New application areas emerge in the ICT domain (simulation of adaptive agent systems, design of intelligent agents, profiling, electronic commerce, data mining) and the social sciences, where for example complex adaptive systems are seen as agent-based systems (economic markets, grouping and collaboration of individuals, etc.). The combination of these areas with software agents yields new research in E-commerce, as an economic sub-area.
The usage of evolutionary techniques in these areas requires substantial research into computational feasibility and complexity of the phenomena in question. Modelling and assessing the relation between computational techniques and real-world mechanisms also become very important. CWI presently has a strong interest in multi-agent systems, as they appear in economics, E-commerce, logistics and social systems. Two new projects were started in these areas. The first, Autonomous Systems of Trade Agents in E-commerce, is carried out in cooperation with the national Organisation for Applied Research (TNO), several major companies, and the Telematics Institute. It addresses the development of learning agents for E-commerce applications, including negotiation, search, logistics, profiling, and filtering. In the project Evolutionary Exploration Systems for Electronic Markets, an explorative system is developed to model and study electronic markets, based on evolutionary methods. The aim is to gain insight into the effects of new mechanisms in electronic commerce. Here CWI cooperates with the Faculty of Economics of the University of Amsterdam.
In the field of dynamic algorithms CWI is involved, jointly with Philips Research Laboratories, in the design of fast, dedicated algorithms for on-line management problems in ICT, for example multimedia (eg, scheduling and resource management problems) and telecommunication (eg, on-line assignment of bandwidth). The continuous decision making in real-time within an active system requires on-line management. Here, the problem field can include agents that negotiate about, eg, the quality of service or, more general, that make on-line decisions.
Using several kinds of neural networks, CWI developed a tool for visualization and clustering of high-dimensional data sets by means of data projection (cooperation with Nijenrode University and the Ministry of Public Works). The tool forms an important part of explorative data analysis within a decision making process. Other research into neural networks concerns their use in clustering techniques to determine which customer characteristics are important for purchasing behaviour. There is growing research interest in spiking neural networks. CWI currently focuses on classification problems, for example unsupervised classification to extract information on disaster areas from remote sensing data (in cooperation with the University of Amsterdam and the Cold Regions Research and Engineering Laboratory in the USA). It is anticipated that neural networks can become an important implementation of intelligent agents for several types of tasks.
Han La Poutré - CWI
Tel: +31 20 592 4082