Modelling Research by Edward Tsang

My research can roughly be grouped into two overlapping areas: constraint satisfaction and computational finance and economics. Modelling plays an important part in both of these areas. It works closely with simulation and machine learning (see brief introduction here). This page summarizes some of my projects in modelling in these two areas.

Constraint Satisfaction Constraint satisfaction is a mature research area. It supports a multi-million Pounds business. Before any constraint techniques can be applied, the problem must be formulated as a constraint satisfaction problem. The Constraint Programming and Optimization Laboratory pioneered modelling research in constraint satisfaction. It attempts to systematically guide the users to identify constraints, domains and constraints in a problem. James Borrett produced the first PhD thesis (1998) in constraints modelling; it reports a mechanism for selecting constraint models. These are small but significant steps in constraints modelling. Modelling has since become a major research topic in constraint satisfaction.
Workforce Management
BT has over 12,000 technicians in UK. They perform tens of thousands of repair and maintenance jobs all over Britain everyday. If we could save them 0.5%, we are talking about savings in terms of millions (Pounds) a year. We have built a model to capture the business processes in engineers management. With this model, we intend to experiment with alternative procedures and management styles, such as staff empowerment.
Card Payment Market Electronic payment is the norm in some countries, but is only starting in others. The card payment market is complex. It requires the willing participation of card issuers, merchants and customers. Models of the card payment market has been proposed, but they were only studied based on observations or small scale experimentation through human subjects. We have implemented a model and studied the equilibrium under different market conditions. We have also used evolutionary computation to evolve strategies for card issuers to maximize their market share or profit.
Artificial Stock Markets By modelling the behaviour of the participants in a particular market, one can attempt to study the market's dynamics and predict market equilibrium. We have identified conditions under which the prices in an artificial market share similar statistical properties as those in real markets. This does not mean that we have fully understood how prices change in a real market. However, it does give us means of better understanding the market mechanism and testing trading strategies.
Related project:
Automated Bargaining
Game theory is an important subject in many disciplines, such as economics, business and sociology. Bargaining situations are modelled both mathematically and procedurally. Traditionally, equilibriums in bargaining are derived mathematically. Such derivations assume that players are fully rational. Unfortunately, this is not always true. Besides, mathematical derivations are often laborious. In order to relax the full rationality assumption, and to study more complex bargaining scenarios, we take an evolutionary approach. Experimental results so far show that evolutionary computation is an effective way to approximate equilibriums in bargaining.

Maintained by Edward Tsang; Last updated 2008.12.15