Learning to describe and efficiently recognize patterns and objects in scenes

Machine learning has been applied to many problems related to scene interpretation. It has become clear from these studies that it is important to develop or choose learning procedures appropriate for the types of data models involved in a given problem formulation. In this paper, we focus on this issue of learning with respect to different data structures and consider, in particular, problems related to the learning of relational structures in visual data. Finally, we discuss problems related to rule evaluation in multi-object complex scenes and introduce some new techniques to solve them.