Hybrid Architecture Based on Support Vector Machines

An hybrid SVM-symbolic architecture for classification tasks is proposed in this work. The learning system relies on a support vector machine (SVM), meanwhile a rule extraction module translate the embedded knowledge in the trained SVM in the form of symbolic rules. The new representation is useful to understand the nature of the problem and its solution. Moreover, a rule insertion module in the hybrid architecture allows incorporate the available prior domain knowledge into the machine expressed in the form of symbolic rules. Thus, it is render possible the integration of SVMs with symbolic AI systems.