Linear Prediction in a Multi-Agent Environment

Joachim Baczewski and Mario Tokoro

In Multi-Agent systems, prediction of future states of observed events in an open and distributed environment gains importance with the growing complexity of the events agents have to deal with. We develop a model for applying Linear Prediction to a Multi-Agent environment. We design a prediction mechanism which senses its environment just by observing plain sampled data. No complex model of the environment has to be built in order to do good predictions of future events. Thus, our agents can react in a real-time manner by simply analyzing the observed events. After introducing the theoretical background of Linear Prediction, we show that Linear Prediction can give good results even for multiple step predictions over a wide range of dynamics. Further, we show that, if agents apply our model competitively in a heterogeneous society of multiple agents, the stability and the performance of the whole system improves.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.