Towards Integration of Different Paradigms in Modeling, Representation and Learning of Visual Categories