Geometrically Guided Integrated Gradients

Md Mahfuzur Rahman, Noah Lewis, Sergey Plis

Interpretability methods for deep neural networks mainly focus on the sensitivity of the class score with respect to the original or perturbed input, usually measured using actual…