Organiser: Dr David Millard
Time: 09/05/2005 12:50-13:40
There is an increasing need for computer systems that operate a decentralised control regime, that are open (individual components can enter and leave at will) and that contain a number of components representing distinct stakeholders with different aims and objectives. Relevant examples include grid computing, the semantic web, pervasive computing, e-commerce, mobile computing and peer-to-peer systems.
For these complex systems, it has been argued that agent-based approaches are a natural computational model. In such multi-agent systems, there are two fundamental design issues that need to be addressed. First, there is a need to specify the protocols that govern the interactions. Second, given the prevailing protocol, there is a need to define the strategy (mapping from state history to action) for each agent.
Game theory provides a formal foundation and has developed powerful tools for addressing these issues. These tools, when tailored to computational settings, provide a foundation for building multi-agent software systems. This tailoring gives rise to the field of computational-mechanism design, which applies economic principles to computer systems design.
In this talk, we outline some of the challenges involved in applying some of the standard tools of game theory to the design of computational systems. These range from finding computationally tractable solutions to the design of utility functions and a common numeraire (i.e. money) in settings where these do not occur. We illustrate these challenges by considering the allocation of bandwidth in a multi-sensor network.
Any member of ECS with their name preceded by a sign has not given their permission for their information to appear on the public website.