Research map / Learning theory and algorithms / Prediction and decision losses
Synthetic Benchmarks for Scientific Research in Explainable Machine Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 SHAPR instead of SHAP | Feature attribution on Gaussian nonlinear additive data · Gaussian nonlinear additive, ρ=0.5 | GT-Shapley: 0.810 ± 0.023 -> 0.930 ± 0.012 | Helped |
| Use BREAKDOWN instead of SHAPR | Feature attribution on Gaussian nonlinear additive data · Gaussian nonlinear additive, ρ=0.5 | monotonicity: 0.550 ± 0.025 -> 0.562 ± 0.021 | Helped |
| Use MAPLE instead of SHAPR | Feature attribution on Gaussian nonlinear additive data · Gaussian nonlinear additive, ρ=0.5 | infidelity: 0.036 ± 0.013 -> 0.019 ± 0.011 | Helped |
| Increase feature correlation | LIME · Gaussian synthetic datasets | faithfulness: ~ 90% | Hurt |
| Increase feature correlation | MAPLE · Gaussian nonlinear additive and piecewise constant datasets | faithfulness | No clear effect |
| Use SHAPR instead of SHAP | Feature attribution on simulated wine · Simulated wine | GT-Shapley: 0.825 ± 0.006 -> 0.945 ± 0.002 | Helped |
| Use SHAP or SHAPR instead of LIME or MAPLE | Feature attribution on simulated wine · Simulated wine | faithfulness | Helped |
| Use SHAP instead of LIME | Feature attribution on simulated forest fire · Simulated forest fire | GT-Shapley: 0.779 ± 0.027 -> 0.870 ± 0.005 | Helped |
Closest papers by meaning
- $\mathcal{M}^4$: A Unified XAI Benchmark for Faithfulness Evaluation of Feature Attribution Methods across Metrics, Modalities and Models (NeurIPS 2023 · similarity 0.89)
- $\texttt{dattri}$: A Library for Efficient Data Attribution (NeurIPS 2024 · similarity 0.86)
- quanda: An Interpretability Toolkit for Training Data Attribution Evaluation (NeurIPS 2026 · similarity 0.86)
- Feature Attribution with Necessity and Sufficiency via Dual-stage Perturbation Test for Causal Explanation (ICML 2024 · similarity 0.86)
- Provably Better Explanations with Optimized Aggregation of Feature Attributions (ICML 2024 · similarity 0.85)
- Local vs. Global Interpretability: A Computational Complexity Perspective (ICML 2024 · similarity 0.85)
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.