AI in Wireless Networks: The Personal Router Agent for Wireless Access

Elicitation of user preferences has been recognized to be one the most important goals of user-centered AI systems. Solutions to this problem have been cast as a utility function construction problem to adaptive classi£cation given classes of utility functions, to sequential decision making. In this paper we present the preference elicitation problems involved in dynamic user access to wireless networks. We propose an interactive, adaptive agentbased solution to the problem and show how the full nature of the problem can be represented within a Markov Decision Process (MDP). Adaptive reinforcement learning solutions are then evaluated for two subclasses of tractable MDPs via simulations of some representative user models.