Locally Linear Embedding

Traditional visualization techniques can hardly be used to analyze very high-dimensional data. Consider, for example, applying the Chernoff-Flury faces of Chap. 1.5 to identify patterns of interest within data that represent images. In this case, the pixel intensity vectors that embody the image can easily have thousands of dimensions. A similar issue occurs in textual analysis where word-count vectors are employed to describe documents. In these situations, Chernoff-Flury faces can no longer be applied. Alternative techniques like Andrews’ curves and parallel coordinate plots will provide such a bad “signal-to-ink” ratio that human interpretation of the graphical representation of the data is impossible.