A symbolic-neural classification system assisting the characterization of the Lyme-disease

The work presented is intended as a step toward hybrid systems and describes a closed-loop integration of a connectionist network into a symbolic concept-learning system. The system is aimed at exploiting many of the powerful features of neural nets to enhance the inherent capabilities of symbolic inductive learning. Backpropagated multilayer perceptrons have been employed to adaptively control how the symbolic system consults declarative knowledge. Conversely, the training of neural networks is under the full control of symbolic tasks. The test case is related to a still open medical diagnosis problem. The main features of the hybrid system are discussed in the light of the preliminary results obtained