Abstract: In this paper the hierarchical approach in learning and its application in building a hybrid (symbolic - subsymbolic) system for emotion recognition is examined. Using the hierarchical approach, instead of creating a solid and direct mapping from a set of inputs to the set of the corresponding outputs, the process is divided into a number of stages where the goal is to construct an abstract description of the concept to be learned. This principle is followed by PAC meditation which is a probably approximate correct (PAC) learning paradigm for Boolean formulas. An extension of this system in order to allow the expression of more complex hypotheses is to incorporate subsymbolic processing at various stages.

Key-Words: PAC learning, PAC meditation, hybrid systems, abstraction learning, emotion recognition