1st Edition

Distributed Strategic Learning for Wireless Engineers

By Hamidou Tembine Copyright 2012
    496 Pages 45 B/W Illustrations
    by CRC Press

    496 Pages 45 B/W Illustrations
    by CRC Press

    Although valued for its ability to allow teams to collaborate and foster coalitional behaviors among the participants, game theory’s application to networking systems is not without challenges. Distributed Strategic Learning for Wireless Engineers illuminates the promise of learning in dynamic games as a tool for analyzing network evolution and underlines the potential pitfalls and difficulties likely to be encountered.

    Establishing the link between several theories, this book demonstrates what is needed to learn strategic interaction in wireless networks under uncertainty, randomness, and time delays. It addresses questions such as:

    • How much information is enough for effective distributed decision making?
    • Is having more information always useful in terms of system performance?
    • What are the individual learning performance bounds under outdated and imperfect measurement?
    • What are the possible dynamics and outcomes if the players adopt different learning patterns?
    • If convergence occurs, what is the convergence time of heterogeneous learning?
    • What are the issues of hybrid learning?
    • How can one develop fast and efficient learning schemes in scenarios where some players have more information than the others?
    • What is the impact of risk-sensitivity in strategic learning systems?
    • How can one construct learning schemes in a dynamic environment in which one of the players do not observe a numerical value of its own-payoffs but only a signal of it?
    • How can one learn "unstable" equilibria and global optima in a fully distributed manner?

    The book provides an explicit description of how players attempt to learn over time about the game and about the behavior of others. It focuses on finite and infinite systems, where the interplay among the individual adjustments undertaken by the different players generates different learning dynamics, heterogeneous learning, risk-sensitive learning, and hybrid dynamics.

    Introduction to Learning in Games
    Basic Elements of Games
    Robust Games in Networks
    Basic Robust Games
    Basics of Robust Cooperative Games
    Distributed Strategic Learning
    Distributed Strategic Learning in Wireless Networks
    Strategy Learning
    Strategy Learning under Perfect Action Monitoring
    Fully Distributed Strategy-Learning
    Stochastic Approximations
    Chapter Review
    Discussions and Open Issues
    Payoff Learning and Dynamics
    Learning Equilibrium Payoffs
    Payoff Dynamics
    Routing Games with Parallel Links
    Numerical Values of Payoffs Are Not Observed
    Model and Notations
    Hybrid and Combined Dynamics
    Learning in Games with Continuous Action Spaces
    CODIPAS for Stable Games with Continuous Action Spaces 183
    CODIPAS-RL via Extremum-Seeking
    Designer and Users in an Hierarchical System
    From Fictitious Play with Inertia to CODIPAS-RL . . . . . 191
    CODIPAS-RL with Random Number of Active Players
    CODIPAS for Multi-Armed Bandit Problems
    CODIPAS and Evolutionary Game Dynamics
    Fastest Learning Algorithms
    Learning under Delayed Measurement
    Learning under Delayed Imperfect Payoffs
    Reacting to the Interference
    Learning in Constrained Robust Games
    Constrained One-Shot Games
    Quality of Experience
    Relevance in QoE and QoS satisfaction
    Satisfaction Levels as Benchmarks
    Satisfactory Solution
    Efficient Satisfactory Solution
    Learning a Satisfactory Solution
    From Nash Equilibrium to Satisfactory Solution
    Mixed and Near-Satisfactory Solution
    CODIPAS with Dynamic Satisfaction Level
    Random Matrix Games
    Mean-Variance Response and Demand Satisfaction
    Learning under Random Updates
    Description of the Random Update Model
    Fully Distributed Learning
    Dynamic Routing Games with Random Traffic
    Mobility-Based Learning in Cognitive Radio Networks
    Hybrid Strategic Learning
    Chapter Review
    Fully Distributed Learning for Global Optima
    Resource Selection Games
    Frequency Selection Games
    User-Centric Network Selection
    Markov Chain Adjustment
    Pareto Optimal Solutions
    Learning in Risk-Sensitive Games
    Risk-Sensitive in Dynamic Environment
    Risk-sensitive CODIPAS
    Risk-Sensitivity in Networking and Communications
    Risk-Sensitive Mean Field Learning
    Chapter Review
    Basics of Dynamical Systems
    Basics of Stochastic Approximations
    Differential Inclusion
    Markovian Noise


    Hamidou Tembine