It is often important to search high dimensional spaces looking for patterns more complicated than maxima or minima of cost functions. For example, we can provide users with visualizations of two of three dimensional slices through such a space, but need to automate finding interesting pictures from among the very large number of possible slices. This problem, which can be called "subspace pursuit", has proven very important for recent work in using exploratory modeling to understand complex systems. Conceptually akin to projection pursuit methods popularized in statistics, subspace pursuit is a problem area that has potential importance for a wide range of applications. In this paper, I present an evolutionary approach to making such searches.
展开▼
