Reproduce “FRInGe: Distribution-Space Integrated Gradients with Fisher--Rao Geometry” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Gabriele Martino, Sebastian Tschiatschek
FRInGe explains vision-model predictions by integrating target-score gradients along an input path planned in prediction space. It starts at the model’s prediction and tracks Fisher–Rao geodesic targets toward a uniform distribution, using pullback Fisher geometry, trust regions, and spatial smoothing. On 1,000 ImageNet images across six architectures, it ranks first in MAS-Insertion on all six; MAS-Deletion ranks first on three, while other metrics are mixed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.