Neurl Network-Based Decision Making for Large Incomplete Database

As an extension to the relational algebra, maybe algebra operations have been proposed to handle incomplete information. Such a set of operations allows the user to investigate the potential set of data values (i.e. tuples) to draw his/her own conclusions. However, maybe algebra operations could return nonrelevant data, generate low quality results, and offer low physical performance. Hence, it is appropriate to design a scheme to investigate the results generated by the maybe operations, in order to improve the data quality and performance of large databases. Such a mechanism should be dynamic to adjust itself according to the user's query and the characteristics of the underlying databases. In this paper, an artificial neural network-based decision support system for handling large databases containing incomplete information is proposed. It is a subsystem which learns and constructs a knowledge base to filter out the data that is not of any importance to the user. The network accomplishes the decision-making task in a massively parallel manner. This paper also discusses the implementation of the decision-making network based on the VLSI design of a Basic Neural Unit (BNU). Using a weight-centered design principle, BNU can be expanded and reconfigured to satisfy the requirements of the underlying environment.