A Cooperation Strategy for Decision-making Agents

The style of engineering computing may change dramatically in the future. Large numbers of agents with local perception of the environment, autonomy, and considerable intelligence will be made to collaborate and form decision-making agents. When agents make correct decisions more often than not with probability p, a cooperation strategy based on simple majority makes the resulting decision-making agent more reliable than the underling agents. However, simple majority does not necessarily improve overall performance if agents have diierent probabilities of being correct. An alternative weight-based cooperation strategy is proposed so as to improve overall performance , and prevent the cooperating agents from being worse than the best member. This paper also tackles the design of such a strategy, derives some theoretical results, and demonstrates the main ideas in an application to face recognition using artiicial neural networks.