Policy Invariance under Reward Transformations for General-Sum Stochastic Games
Computer Science and Game Theory
2014-01-17 v1 Machine Learning
Abstract
We extend the potential-based shaping method from Markov decision processes to multi-player general-sum stochastic games. We prove that the Nash equilibria in a stochastic game remains unchanged after potential-based shaping is applied to the environment. The property of policy invariance provides a possible way of speeding convergence when learning to play a stochastic game.
Keywords
Cite
@article{arxiv.1401.3907,
title = {Policy Invariance under Reward Transformations for General-Sum Stochastic Games},
author = {Xiaosong Lu and Howard M. Schwartz and Sidney N. Givigi},
journal= {arXiv preprint arXiv:1401.3907},
year = {2014}
}