Modelling derivation in defeasible logic programming with perceptron-based neural networks

Abstract

A solution of problems in multiagent systems involves representing beliefs of agents immersed in dynamic environments. Observation-based Defeasible Logic Programming (ODeLP) is an argument-based logic programming language that is used to represent an agent’s knowledge in the context of a multiagent system. The beliefs of the agent depends on a warrant procedure performed on its knowledge base contents. New perceptions result in changes in the agent’s beliefs. In the context of real time constraints, this belief change procedure should be done efficiently. This paper introduces an algorithm for translating an agent’s knowledge base, expressed as an ODeLP rule base, into a Perceptron-based neural network. Observations in an ODeLP program can then be codified as an input pattern. The input pattern is then fed to the neural network whose propagation results in an output pattern. This output pattern contains information regarding which beliefs can be hold by the agent as well as if there exists contradiction among them. The proposal is attractive as the massivelly parallel processing intrinsic to neural networks make them appropiate for implementing parts of the aforementioned warrant procedure.