In many fields of our society, experts are called upon to solve complex decision problems in their every-day problem-solving practice. Examples include the medical field where specialists have to establish diagnoses and decide upon appropriate treatment alternatives, and the financial field where companies have to forecast demand and decide upon investments. In many of these fields, the problems to be solved are rapidly increasing in complexity, and highly tailored support from computer-based systems for decision making is called for. The aim of the Decision-Support Systems research programme is to design a new generation of decision-support systems that are capable of handling the increasing complexity of the problems in their domains of application.
The Decision-Support Systems research programme is aimed at the design and analysis of decision-support systems that build upon concepts and techniques from statistics. Decision-support systems nowadays often call for the use of such concepts and techniques, either because the problems to be solved in their domain of application require reasoning with uncertainty for their solution or because the complexity of these problems forestalls the use of exact algorithms. Within the programme, the use of concepts and techniques from statistics for decision support is being studied in two main research themes:
- decision making under uncertainty
- evolutionary computation
The research themes are being studied mainly from a fundamental computer-science perspective, focusing on representation formalisms and associated algorithms. Although firmly rooted in the field of statistics, statistics itself is not the focus of research. The viability and practicability of the more fundamental research results are being studied in experimental settings and in real-life applications that are being developed with the help of domain experts.
Decision making under uncertainty
Decision making under uncertainty is the field of research that addresses reasoning with uncertainty and reasoning about preferences for solving complex decision problems. While traditionally building upon techniques from statistics, the field is now showing an increasing impact from computer science. Research efforts have resulted in powerful formalisms for capturing the probabilistic and preferential relationships between a decision problem’s variables. These formalisms typically build upon graphical representations, such as probabilistic networks and decision-theoretic networks more in general. Algorithms have been designed for computing probabilities of interest and optimal decisions from such networks.
Within the research programme, the practicability of the above formalisms and their associated algorithms is the main focus of interest. The design of methodologies and techniques for constructing and analysing probabilistic networks is being pursued, as well as the design of efficient algorithms for various types of probabilistic reasoning. To provide for studying the more fundamental research results in a realistic setting, a number of real-life probabilistic networks are being developed for complex medical (both human and veterinary) decision problems.
Evolutionary computation is the field of research that studies algorithms for search, optimisation, and adaptation that are inspired by the mechanisms of natural evolution and genetics. Neo-darwinian principles can lead to efficient and robust search mechanisms as exemplified by the evolution of species, the adaptive recognition abilities of the human immune system, and the self-organising development of neural pathways. In the field of evolutionary computation, the general computational principles of these mechanisms are taken for the basis of powerful algorithms for problem solving.
Within the research programme, the characteristics of evolutionary search are being studied. In addition, the development of methodological guidelines for designing evolutionary algorithms as well as the design of new types of evolutionary algorithm based upon new principles are being pursued. Research activities also include the use of genetic algorithms and related stochastic search algorithms for constructing probabilistic networks from data.
The current Decision-Support Systems programme was started in May 2000, through the appointment of L.C. van der Gaag as a full professor, upon being awarded the prestigious five-year pioneer-funding from the Netherlands Organization for Scientific Research (NWO Pionier) for her research programme Practicable Decision Making Under Uncertainty. In 2007, professor Van der Gaag was honoured with an ECCAI Fellowship.