Home

About

Advanced Search

Browse by Discipline

Scientific Societies

E-print Alerts

Add E-prints

E-print Network
FAQHELPSITE MAPCONTACT US


  Advanced Search  

 
On Non-Cooperative Interaction via Reinforcement Learning and Its Control
 

Summary: On Non-Cooperative Interaction via Reinforcement
Learning and Its Control
Murat Alanyali
Department of Electrical and Computer Engineering
Boston University
alanyali@bu.edu
Abstract
A general setting is considered in which an arbitrary set of autonomous users
interact by means of a controlled Markov process. This process is driven by the col-
lective actions of users, and users receive separate payoffs according to its state. The
only other assumption on the process is that it has a finite state space. Dynamic
behavior of user strategies is characterized in the case when each user exercises a
reinforcement learning algorithm based on its local information. This characteri-
zation leads to pricing guidelines for determining payoffs as a decentralized control
mechanism. This principle is applied to routing and flow control in packet data net-
works, and practically appealing definitions of payoffs are shown to lead to socially
efficient network operation.
1 Introduction
This paper characterizes dynamics of non-cooperative interaction when individuals exer-
cise reinforcement learning in choosing their actions. The main motivation of the paper

  

Source: Alanyali, Murat - Department of Electrical and Computer Engineering, Boston University

 

Collections: Engineering