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