For the past few years we have been working on various empirical approaches to mechanism design. This is not the automated mechanism design of Sandholm and Conitzer, which seeks to generate provable correct but simple mechanisms, but work that aims to optimise aspects of complex mechanisms, chiefly double auctions. I will describe a number of approaches based on machine learning, and show the results we have obtained for single auctions. I will also describe more recent work exploring the properties of multiple, interacting auctions, and examine results from the TAC CAT game, which takes this form.
Dr Simon Parsons
Simon Parsons is a Professor of Computer Science at Brooklyn College of the City University of New York (CUNY). Simon got his PhD in 1993 from the University of London, and took his first faculty job two years later. He worked at the University of London, the University of Liverpool and MIT before moving to CUNY in 2002. Simon has over 200 publications, many on the subject of argumentation in multiagent systems. In recent years his main research interest has been focused on mechanism design.