Discovering Graph Structures In HighDimensional Spaces

This paper introduces a new idea to face this problem by adding the notion of functional dependency to the context of minimum encoding inference. Our analysis is based on a local discovery of relations between binary features. We present a very effective expression of class model local note, wich can thus be computed without knowledge of any parameter value. The first method of data analysis described in this paper (also called crude method) is the direct application of our expression of the model class local note, and produce pieces of knowledge discovered in the data, organized in a graph.