AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

Deletion and Insertion Tests in Regression Models

Naofumi Hama, Masayoshi Mase, Art B. Owen

JMLR 2023 · Learning theory and algorithms › Prediction and decision losses · approach: Provable post-hoc explanations · 9 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use Kernel SHAP rather than Integrated Gradients for feature orderinginsertion test · Bangalore housinginsertion ABC: 0.628 vs 0.572Improved
use Kernel SHAP rather than Integrated Gradientsdeletion test · Bangalore housingdeletion ABC: 0.423 vs 0.422No clear effect
use vanilla gradient instead of random orderingfeature ranking · Bangalore housinginsertion ABC: -0.093 vs -0.020Hurt
use Kernel SHAP rather than Integrated Gradientsfeature ranking · CERN counterfactual policyinsertion ABC: 18.535 vs 18.289Improved
use Kernel SHAP rather than Integrated Gradientsfeature ranking · CERN counterfactual policydeletion ABC: 16.752 vs 16.315Improved
switch reference from counterfactual to average policy for LIMELIME · CERNinsertion ABC: 17.319 -> 11.892Hurt
use Kernel SHAP rather than Integrated Gradientsfeature ranking · CERN one-to-one policyinsertion ABC: 30.392 vs 28.430Improved
use Kernel SHAP rather than Integrated Gradientsfeature ranking · CERN average policydeletion ABC: 7.621 vs 6.856Improved
change reference policy from counterfactual to one-to-oneinsertion/deletion correlation (KS) · CERNcorrelation of insertion and deletion ABC: 0.820 -> -0.276Hurt
change reference policy from counterfactual to averageinsertion/deletion correlation (KS) · CERNcorrelation of insertion and deletion ABC: 0.820 -> 0.587Hurt
use a blurred rather than black reference imagedeletion test · ImageNet albatross exampledeletion AUC: 0.270 -> 0.981Hurt
use a mean rather than black reference imageinsertion test · ImageNet albatross exampleinsertion AUC: 0.201 -> 0.187No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy