Develop software tools to support the techniques; Extend and further develop the techniques themselves; Establish a methodology for using the techniques; and Test the techniques on a pair of applications.
Auctions are becoming increasingly common, both in the award of government contracts and in commercial transactions. Although on-line auction houses such as e-Bay obtain perhaps more publicity, electronic business-to-business auctions now account for the majority of e-commerce transactions. Traditionally, game theory has been used by economists to design auction mechanisms, but, for various reasons, these undoubtedly powerful analytic methods are not always successful or appropriate. We believe a computational approach is required, and in our prior work we have explored, in successful proof-of-concept research, two techniques for computational auction mechanism design. One of these is a co-evolutionary approach, in which mechanisms and participant trading strategies evolve together over time. The other uses reinforcement learning to set trading strategies and then searches for the best mechanism given these strategies.
In this project, we seek to take these two techniques beyond the proof-of-concept stage. Our key aims are to further develop the two techniques outlined above, and software tools to support them, towards a point at which they can assist those who design auctions---possibly in combination with traditional analytic techniques---in the same way that computer aided design (CAD) packages help mechanical engineers and architects. To do this we will
Develop software tools to support the techniques; Extend and further develop the techniques themselves; Establish a methodology for using the techniques; and Test the techniques on a pair of applications.
(Taken from the project proposal...)
More specific information about the work carried out on the project may be found in the papers on the publications page.