Anticipatory Networks

Abstract

This chapter presents a network structure for a class of anticipatory systems in Rosen’s sense linked by causal and information transfer relations. They form a multigraph termed an anticipatory network (AN). The ANs were first defined in order to model a compromise solution selection process in multicriteria optimization problems. The nodes in such a network, termed optimizers, are capable of selecting a nondominated solution taking into account the anticipated consequences of a decision to be made. Specifically, to make a decision in a problem associated to an optimizer, the decision-maker should take into account the anticipated outcomes of each future decision problem linked by a causal relation with the current one. The ANs presented in this chapter are based on an assumption that constraints and preference structures in nodes associated to future decision problems may depend on the values of criteria that result from solving preceding problems. The nodes of a hybrid AN may correspond to optimizers, random, conflicting, interactive, or predetermined (nonautonomous) decision problems. We will overview most relevant types of ANs, their solution concepts, and constructive solution algorithms. It will be pointed out that the structure of an anticipatory network imposes the superanticipatory property for its components. We will also present timed AN as well as further extensions of ANs with various information exchange relations. The discussion section contains a survey of reallife applications, including coordination of autonomous robotic formations, foresight scenario filtering, strategy building, and others.