A DECENTRALIZED OPTIMAL SOLUTION TO THE “ MINORITY GAME ” by Thiagarajan

The minority game is relevant to current large scale information systems (such as the internet) as a model of the distributed nature of decision making in situations where agents must compete for a finite resource, e.g., bandwidth. This thesis reviews the minority game and discusses a “socially” optimal adaptive solution using a adaptive mixed strategy (i.e., stochastic) approach and a partial information structure. The algorithm has a formal similarity to adaptive signal processing algorithms and it is simple to implement. The algorithm is analyzed using a weak convergence approach that identifies (locally) stable equilibrium states to which the algorithm converges. Variations on the algorithm, based on other information structures are explored, and the thesis compares and contrasts the different schemes, both analytically and with relevant simulations.