NETLEARN: Orchestration d’algorithmes d’apprentissage distribués pour la gestion des ressources dans les réseaux mobiles
Project Presentation
NETLEARN is a collaborative research project financed by the french national agency for research (ANR) and which aims at developing new tools based on recent advances in learning theory for the management of various resources in mobile networks. It started the 21 Oct. 2013 and will end the 20 Apr. 2017. Partners of the projects are: INRIA (LIG), UVSQ (PRISM), University Paris Dauphine (LAMSADE), Institut Mines-Telecom (Telecom ParisTech), Alcatel-Lucent Bell Labs and Orange Labs. It is organized in four work packages.
See the project's web page NETLEARN
software library
This work was written to help in the use of the Game Theory Framework application. The Game Theory Framework application focuses on providing simulations of games, with players simulated by learning algorithms implemented in the application itself. It allows to play out scenarios which would be complicated to visualize theoretically, mostly in the case of multi-player scenarios, but also when the players or/and actions for each player are numerous. The aim of this help is to explain the underlying structure used in the framework, as well as the way games and learning algorithms should be understood and expanded upon.
The implemented learning algorithms are the follows : BestResponseAverage, variants of UCB, Hedge, EXP3, FTL. Also, There are two kind of the implemented games : Two player Game, Congestion Game This library has been implemented in C++ by F. Dufulon.