DMCA
Learning Optimal Control in Deterministic Systems
BibTeX
@MISC{For_learningoptimal,
author = {Learning Algorithms For and Pareigis Stephan},
title = {Learning Optimal Control in Deterministic Systems},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Introduction In some optimal control problems a solution cannot be obtained by standard numerical methods. This may be due to very large state spaces (as in game playing) or incomplete information such as unknown system dynamics. The mathematical model of the problem may also be either too complicated to handle or too simple to be accurate enough. A promising method is to learn the optimal value function by letting the real system (or simulation) perform the dynamics. The optimal value function is then approximated using the information provided by the system: state, local cost and local control. This computation may be off-line as in game playing (the computer gains experience by playing against himself), or on-line as in heavy traffic problems (the computer learns about the dynamics of a data network by actually controlling it). 2. Formulation of the learning problem We want to approximate the optimal value function V :\Omega !<F12.3
Keyphrases
optimal value function deterministic system optimal control game playing promising method local control omega f12 data network heavy traffic problem mathematical model solution cannot local cost optimal control problem incomplete information standard numerical method unknown system dynamic computer gain experience learning problem real system large state space
