Interpolative Reasoning in Distributed Intelligent Systems Based on Fuzzy Sets and Neural Nets

Abstract

Interpolative reasoning is an inference scheme proposed by Professor Loth A. Zadeh to cope with imprecise and incomplete information in an inference process. It is closely related to fuzzy modeling , pattern classification , and learning from examples . The purpose of this paper is to provide an interpolative reasoning mechanism realized by a generalized neural network . We pay special attention to the needs of a distributed system in our design, such as weighted matching and adaptive learning. Our approach in this paper begins with identifying the issues of reasoning in a distributed system. Then, we introduce the concept of generalized neural networks (GNN’s) and derive a supervised learning procedure based on a gradient descent algorithm to update the parameters in a GNN. Next, we apply the GNN architecture to implement a fuzzy inference system with weights of importance. We also enhance the basic model by using fuzzy quantifiers. An example of simulation result is illustrated. Finally, directions of future work are suggested.