Learning Equilibria with Partial Information in Decentralized Wireless Networks
Abstract
In this article, a survey of most suitable equilibrium concepts for decentralized self-configuring networks is presented. Here, the term decentralized is used to refer to scenarios where decisions (e.g choosing a power allocation policy) are taken autonomously by interactive radio devices. The iterative long term interaction is characterized by stable points of the wireless network called equilibria. The interest for these equilibria basically stems from the relevance of stability, while the interest for decentralization comes along with network scalability and flexibility. In order to achieve the equilibria mentioned above, several learning techniques (namely: best response dynamics, fictitious play, smoothed fictitious play, cumulative reinforcement learning, JUSTE reinforcement learning and regret matching) are introduced and analysed in terms of information assumptions and convergence properties. Most of the notions introduced here, both equilibria and learning schemes, are illustrated by a simple study case, namely, the interference channel with two transmitter-receiver pairs and two frequency bands.
Origin : Files produced by the author(s)
Loading...