Methods of Automated Rule Discovery in a Multi- agent Adversarial Distributed Environment

摘要

For many applications it is desirable to construct expert systems for decision support. It is often essential that the algorithm make decisions at least at the quality level of the best human experts according to some measure. ft should not only incorporate human expertise and its related uncertainty, but also be able to exploit an opponent's uncertainty. In many instances, due to newly emerging paradigms, the necessary human expertise does not exist. An approach based on fuzzy logic and evolutionary algorithms has been developed that allows the automatic inversion of rules from physics. The rules discovered are shown to reproduce rules already developed by human experts or in the case, where no human being has expertise, to be original. Two approaches to automatic rule discovery are discussed. One of these approaches uses game theory. Along with the theoretical basis, an application to electronic warfare, and related experimental results are given.