Independent components of expertise

This paper presents some results that have been found when a new view of human expertise is employed: it is assumed that the “chunks” of expert knowledge are based on “independent components” in the high-dimensional observation space. These components are exploited by the optimised expert model because of their power of representing the data; and because of their relevance they can also be detected and extracted using automatic statistical tools. As an application example, modelling of the visual outlook of the so called flotation process are presented: it seems that the extracted froth prototypes match the sub-symbolic categorization of the human process operator.