Tag
This paper introduces Fisher width and inverse-Fisher width on statistical manifolds, studying their roles in local learning bounds and anisotropic recovery. It proves a complementary relation between the two widths and obtains recovery estimates based on Fisher geometry.