Contour map drawn from a two‐dimensional (2‐D) array of sums of squares and cross products supposed to represent an estimated, nonseparable variance–covariance matrix.

90% prediction regions constructed from unimodal regression (pink area in the left panel) and multimodal regression (light blue area in the right panel). Clearly, the prediction region is much smaller in the multimodal regression than in unimodal regression because multimodal regression detects all components whereas unimodal regression discovers only the main component.

A proposed framework for constructing SVMs when the training set includes observations with complete predictor information (filled circles) and observations with incomplete predictor information (empty circles).