No Regret Routing for ad-hoc wireless networks

In this paper we consider the problem of adaptive routing in wireless ad-hoc networks with respect to the regret (or learning loss) criterion. The regret criterion accounts for the loss in performance incurred because of the implicit on-line learning task involved when the link probabilities are unknown. In particular, we construct asymptotically efficient adaptive control schemes (that minimize the rate at which regret accumulates with time) that come arbitrarily close to the minimum rate. The intention is to capture the conflict between learning and control.