Game Theory and Learning for Wireless Networks

Chapter 5. Partially Distributed Learning Algorithms

The learning approach to game theory studies what kind of equilibria might occur in a repeated process of learning, adaptation, or imitation. Learning processes are interactive, and to allow them, some of the more rigid assumptions of game theory must be relaxed.

Different types of learning situations are explored. Best response dynamics have been used in wireless applications, in which a player updates his strategy in response to the actions of another. Fictitious play is another learning model used extensively in learning theory to better understand reinforcement-type learning algorithms. Logit or Bolztmann-Gibbs learning has a distribution of actions or strategies that are updated according ...