Research map / Learning theory and algorithms / Prediction and decision losses
Deletion and Insertion Tests in Regression Models
Naofumi Hama, Masayoshi Mase, Art B. Owen
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use Kernel SHAP rather than Integrated Gradients for feature ordering | insertion test · Bangalore housing | insertion ABC: 0.628 vs 0.572 | Improved |
| use Kernel SHAP rather than Integrated Gradients | deletion test · Bangalore housing | deletion ABC: 0.423 vs 0.422 | No clear effect |
| use vanilla gradient instead of random ordering | feature ranking · Bangalore housing | insertion ABC: -0.093 vs -0.020 | Hurt |
| use Kernel SHAP rather than Integrated Gradients | feature ranking · CERN counterfactual policy | insertion ABC: 18.535 vs 18.289 | Improved |
| use Kernel SHAP rather than Integrated Gradients | feature ranking · CERN counterfactual policy | deletion ABC: 16.752 vs 16.315 | Improved |
| switch reference from counterfactual to average policy for LIME | LIME · CERN | insertion ABC: 17.319 -> 11.892 | Hurt |
| use Kernel SHAP rather than Integrated Gradients | feature ranking · CERN one-to-one policy | insertion ABC: 30.392 vs 28.430 | Improved |
| use Kernel SHAP rather than Integrated Gradients | feature ranking · CERN average policy | deletion ABC: 7.621 vs 6.856 | Improved |
| change reference policy from counterfactual to one-to-one | insertion/deletion correlation (KS) · CERN | correlation of insertion and deletion ABC: 0.820 -> -0.276 | Hurt |
| change reference policy from counterfactual to average | insertion/deletion correlation (KS) · CERN | correlation of insertion and deletion ABC: 0.820 -> 0.587 | Hurt |
| use a blurred rather than black reference image | deletion test · ImageNet albatross example | deletion AUC: 0.270 -> 0.981 | Hurt |
| use a mean rather than black reference image | insertion test · ImageNet albatross example | insertion AUC: 0.201 -> 0.187 | No clear effect |
Closest papers by meaning
- Benchmarking Deletion Metrics with the Principled Explanations (ICML 2024 · similarity 0.86)
- What Does It Take to Build a Performant Selective Classifier? (NeurIPS 2025 · similarity 0.83)
- Additive Models Explained: A Computational Complexity Approach (NeurIPS 2025 · similarity 0.82)
- Random features models: a way to study the success of naive imputation (ICML 2024 · similarity 0.82)
- Outlier Gradient Analysis: Efficiently Identifying Detrimental Training Samples for Deep Learning Models (ICML 2025 · similarity 0.82)
- Integrated Decision Gradients: Compute Your Attributions Where the Model Makes Its Decision (AAAI 2024 · similarity 0.82)
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.
