A Temporal-Difference Prediction Approach for Multiplayer Games

T.-W. Yue and M.-P. Hsieh (Taiwan)

Keywords

Reinforcement Learning, TDNetworks.

Abstract

In this paper, we propose a novel learning method in an in formation imperfect environment. The card game “Hearts” is our test target. It is a typical example of imperfect in formation games, and it is so difficult that the traditional reinforcement learning methods can’t learn well. In this ap proach, the agent learns the card playing skill by improving the predictive precision for the probabilities of the occur rences of future events on a developing card game based only on the information available to the agent. The sophis ticated agent, as a result, is able to estimate the opponents’ violent attacks, and make a strong resistance to fight them back. Playing 100 games with MSHEARTS in Microsoft Windows, our well-trained agent won the championship.

Important Links:



Go Back