Linear subspaces
Classification is still expensive
Must either search (e.g., nearest neighbors) or store large PDF¡¯s
Suppose the data points are arranged as above?
Idea—fit a line, classifier measures distance to line