Integrating RBF networks with domain knowledge
Kenneth McGarry, John MacIntyre, Dale Addison · 2002
Often in real-world situations no actual data is available but a domain expert may have a good idea of what to expect in terms of input and output parameter values. If the expert can express these relationships in the form of rules, this would provide a resource too valuable to ignore. The authors illustrate how a single domain rule can be used to synthesize new RBF hidden unit parameters. These new hidden units are incorporated into the original neural network to enable classification on examples the network would otherwise been unable to recognise. The authors then describe a more complex application using fuzzy logic to insert domain knowledge into the RBF network. Fuzzy logic is particularly suited to manage the imprecision and vagueness of natural language. The fuzzy rules are used to synthesize new RBF hidden unit parameters for incorporation into a new or existing network.