Figure 5.

Results for bimodal categories made of two Gaussians of variance $\nu =5$ separated by a distance $\delta$ ranging from 0 to 30. We used $\alpha =0.0005$. (top) probability mass functions of the categories in the 7 cases; (middle) similarity width for all members of a category. As in the uniform and Gaussian category cases, the similarity is wider at the center than at the border; (bottom) goodness of all members of a category. The goodness has a more complicated behavior than in previous cases—from bell-shaped in the $\delta =0$ case (i.e., Gaussian monomodal) to multimodal for large $\delta$.