ENGINEERING MULTIVERSION NEURAL-NET SYSTEMS

In this paper we address the problem of constructing reliable neural-n et implementations, given the assumption that any particular implement ation will not be totally correct. The approach taken in this paper is to organize the inevitable errors so as to minimize their impact in t he context of a multiversion system, i.e., the system functionality is reproduced in multiple versions, which together will constitute the n eural-net system. The unique characteristics of neural computing are e xploited in order to engineer reliable systems in the form of diverse, multiversion systems that are used together with a ''decision strateg y'' (such as majority vote). Theoretical notions of ''methodological d iversity'' contributing to the improvement of system performance are i mplemented and tested. An important aspect of the engineering of an op timal system is to overproduce the components and then choose an optim al subset. Three general techniques for choosing final system componen ts are implemented and evaluated. Several different approaches to the effective engineering of complex multiversion systems designs are real ized and evaluated to determine overall reliability as well as reliabi lity of the overall system in comparison to the lesser reliability of component substructures.