A binary relation inference network Part 1. General concepts

Abstract

Many human and machine reasoning tasks require complicated inferences between objects and events, in which the constituting inference processes depend in turn on successive inferences on more basic binary relations. Given a set of n binary relations between m different objects or events, it is possible to infer other consistent binary relations, to check for relation inconsistency, to resolve conflicts in multiple inferences, by an efficient form of parallel computation: a binary relation inference network. This paper proposes a synchronous computational mechanism for such an inference network, and discusses its topology and physical implementation structures. Network properties and behaviours have also been studied, and some interesting results on computational passes, structural graph, unconstrained and constrained networks, energy functions and convergence conditions are obtained. Potential applications of the inference network for a time-referencing problem and for an autonomous air-traffic controller are technically feasible.