Fig. 1.

Gaussian data with unit variance shown along three axes. The marginal distribution of ${\overrightarrow{e}}_1$ contains signature of distinct clusters, with a bimodal marginal distribution where each mode corresponds to a cluster. Here, the clusters are linearly separable along the e₁ axis. The marginal distributions of ${\overrightarrow{e}}_2$ and ${\overrightarrow{e}}_3$ are unimodal, and do not linearly separate groups of data points. Here, we designate ${\overrightarrow{e}}_1$ as part of V_s as it contains multimodal signal, and $e_2,e_3\in V_n$ do not. The three axes shown here represent a subspace of a D > 3-dimensional distribution