Fisher Information Geometry

The Fisher information matrix g_ij = E[∂_i log p · ∂_j log p] defines a Riemannian metric on the manifold of probability distributions. For Gaussians, this is the hyperbolic metric; geodesics are "information-optimal" paths. The KL divergence is the dual connection's curvature.

Distribution

Fisher det:
KL from ref:
Ricci scalar: