The Representation of Knowledge and Rules in Hierarchical Neural Networks
Neural networks display the property of generalisation. This makes them insensitive to noise and corruption in pattern classification applications. In this chapter we discuss how this property may be exploited in neural network architectures developed for knowledge representation. We show how neural network architectures may be designed to exhibit a hierarchical data-structure suitable for knowledge representation. We also describe how structured rules might be captured within a neural architecture and processed in a manner that can accommodate uncertainty or ambiguity in a reasoning process.