The potential of fuzzy neural networks in the realization of approximate reasoning engines

Abstract

Fuzzy neurons and fuzzy neural networks (FNN) are constructs of computational intelligence that come with significant learning abilities and inherent transparency (interpretability). Consequently, they exhibit high potential to provide strong mechanisms for building intelligent systems that must operate in rapidly changing environments. However, to fully exploit the potential of FNN structures, efficient parallel-processing implementations are highly desired. In this study, our objective is to investigate this avenue and identify various critical hardware design issues as we propose a versatile neurofuzzy platform with a topology strongly influenced by theories of fuzzy modelling. With a novel hybrid-learning scheme involving structural and parametric optimization, we demonstrate how fuzzy neural networks are well suited in forming the adaptive logic-processing core of this platform, supporting intelligent information processing. The core emulates aspects of human thought, dealing with information at a granular level and using logic-oriented approximate reasoning to solve a problem. It is able to learn and approximate real-world concepts, building a knowledge base that may be interpreted and modified by the user. Drawing upon this knowledge, a high-speed parallel implementation has potential for performing inference of many simultaneous concepts in real-time, realizing cognition as it perceives the current state of its environment.