The intersection of Computer Science and Economics has become increasingly important to the development of both fields. Today's software often must handle multiple individuals with their own interests in mind, bringing incentive issues to the forefront in algorithm design. Economic problems, especially in electronic commerce, increasingly involve large numbers of goods and buyers as well as unknown and complex market conditions, making algorithms and machine learning of key importance. This project aims to address fundamental questions at the heart of the intersection of these two fields. These include problems of modeling and influencing behavior in systems with large numbers of agents and components, problems of optimization under complex and changing preferences and constraints in electronic commerce, and problems of efficiently computing and estimating basic economic quantities.
This project specifically has three main thrusts. The first is development of algorithms and analysis techniques for positively influencing dynamics in systems with large numbers of interacting agents. For example, if behavior is currently at a poor-quality equilibrium, when can additional information or a few targeted incentives be used to "nudge" behavior towards a good equilibrium? This applies not only to self-interested agents but also to components in a distributed system acting on local information (such as sensors in a sensor network). The second thrust is development of algorithms for efficiently computing or estimating important economic quantities. This includes approximately computing Nash equilibria in large interactions, and learning submodular functions and other common valuation classes from observations of behavior or experimentation. The third thrust is developing mathematical frameworks for understanding and solving problems of pricing and resource allocation in settings with unknown and changing market conditions. These frameworks are crucial for next-generation markets for resources such as computing power and network bandwidth.
