FIGURE 3 |.

The geometry of model space. In this figure, f₂ and f₃ are approximating models residing in a (hyper)plane. g is the generating model. m is the projection of g onto the (hyper)plane. d(;)^· are distances between models in the plane. d(f₂, f₃) ≈ KL(f₂, f₃) with deviations due to the dimension reduction in NMDS and non-Euclidian behavior of KL divergences. As KL divergences decrease, they become increasingly Euclidian. (A) Shows a projection when m is within the convex hull of the approximating models, and (B) shows a projection when m is outside of the convex hull. Prasanta S. Bandyopadhyay, Gordon Brittan Jr., Mark L. Taper, Belief, Evidence, and Uncertainty. Problems of Epistemic Inference, published 2016 Springer International Publisher, reproduced with permission of Springer Nature Customer Service Center.