Integrating inference and neural classification in a hybrid system for recognition tasks

While the coupling of artificial of neural networks (ANN) and symbolic AI (SAI) is a strategy adopted in many hybrid systems, a real integration of the two methodologies has not been thoroughly investigated yet: so far, most hybrid systems have been viewed as just an engineering shortcut to solve complex problems in which one methodology alone seems too weak.
In this paper, an approach to integrating ANN and SAI is presented. The basic idea explored here is that there is much more to hybrid models than a coupling of methodologies for the sake of avoiding engineering problems.