Riemannian Integrated Gradients: A Geometric View of Explainable AI

We introduce Riemannian Integrated Gradients (RIG); an extension of Integrated Gradients (IG) to Riemannian manif olds. We demonstrate that RIG restricts to IG when the Riemannian man ifold is Euclidean space. We show that feature attribution can be p hrased as an eigenvalue problem where attributions correspond to eig envalues of a symmetric endomorphism.