HANNA: Hybrid architecture for artificial neural network applications

Building applications based on artificial neural networks requires expertise in the domain of neural networks design and utilization as well as in the domain of the application itself. Many promising neural network-based application projects fail to fulfill their objectives because one of these two required kinds of knowledge is missing. To tackle this problem, we have constructed a hybrid architecture consisting of a core neural network simulator tightly coupled with knowledge-based systems. The user interactively defines the domain, problem and goal, while the software progressively refines the task using feedback, questions and data analysis. This process, while not requiring any knowledge of neural networks, can help rite user realize what constitutes meaningful data in their domain.