Multivalued Logic Mapping of Neurons in Feedforward Networks

A common view of feedforward neural networks is that of a black box since the knowledge embedded in the connection weights of a feedforward neural network is generally considered incomprehensible. Many researchers have addressed this deficiency of neural networks by suggesting schemes to obtain a Boolean logic representation for the output of a neuron based on its connection weights. However, these schemes mostly assume binary inputs to the neural network. Since it is not uncommon to find multivalued discrete inputs to neurons, we present in this paper a weight mapping scheme that is capable of generating a multivalued logic representation for the output of a neuron. Such a logic representation is also useful for continuous inputs through multilevel quantization. Two examples are presented to illustrate the use of multivalued logic representation in understanding the knowledge incorporated in the connection strengths of neurons in feedforward networks. Content Areas: Knowledge Acquisition, Machine Learning, Neural Networks. Word Count: 5756 including references. Tracking Number: A199