ABSTRACT
Hybrid systems integrating symbolic and connectionist components should profit from the advantages and strengths of both approaches. The strength of classical symbolic AI lies in deductive reasoning (e.g., expert systems, planning systems, deductive data bases, theorem provers, logic programming). Symbolic systems are based on well-defined and sound logical calculi and allow deep inference chains. This seems infeasible for connectionist systems. Connectionism, on the other hand, offers very successful approaches for inductive learning and generalization and a well-defined concept of similarity (distance in real-valued vector spaces).
