From Game Theory to Distributed Control and Back Again

Thursday, November 6, 2003 - 10:10am - 10:55am
Keller 3-180
David Wolpert (NASA Ames Research Center)
Product Distribution (PD) theory is a powerful new formalism with potential applications throughout science. This talk starts by presenting PD theory in the context of conventional, full-rationality game theory. PD theory is then used to extend conventional game theory, by deriving the information-theoretic formulation of bounded rational game theory. Next the close relationship of that formulation with the canonical ensemble of statistical physics is delineated. This in turn leads to the application of PD theory to distributed control of multi-agent systems. As the audience prefers and time allows, the talk will end with a discussion of PD theory applied to distributed optimization, evolutionary game theory, numerical integration, sampling of probability densities, or management theory.