7. Structural versus statistical approach

Many researchers contrast statistical and structural approaches to pattern recognition. Statistical RDF is concerned with simple objects represented by real feature vectors, without taking structural information into account. Structural RDF, on the other hand, is concerned with complex, structured objects.

Advocates of the statistical approach point to the theoretical framework within which the approach is embedded, which validates its concepts and techniques. Those in favor of the structural approach consider that it enables us (more naturally) to deal with structured objects involving complex relationships between sub-parts, unlike the statistical approach which traditionally exploits "flat" (vector) representations.

This clear-cut view is outdated. For many years, statistical RDF has been concerned with the processing, classification...