Feed-forward neural architectures for membership estimation and fuzzy classification

This paper presents a new decision-making tool called the Quantum Neural Network (QNN), which is capable of autonomously detecting the presence of uncertainty in sample data and adaptively learning to quantify this uncertainty. The QNN is designed by combining the function approximation ability of feed-forward neural networks (FFNNs) with fuzzy-theoretic principles. The advantages of the QNN over other fuzzy classification and inferencing techniques include the ability to approximate any membership profile arbitrarily well from sample data without requiring such restricting assumptions as the availability of a priori knowledge of a `desired membership' profile, convexity of classes, a limited number of classes, etc. Experimental results are presented to demonstrate that QNNs have an inherent ability for recognizing structures in data that conventional FFNNs significantly lack.