Research map / Learning theory and algorithms / Prediction and decision losses
Theoretical Behavior of XAI Methods in the Presence of Suppressor Variables
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 gradient weights to assign feature importance | Bayes optimal linear classifier · Two-dimensional Gaussian binary classification | suppressor-feature importance: w_2 = -\alpha c s_1/s_2 | Hurt |
| Use linear activation patterns instead of model weights | Bayes optimal linear classifier · Two-dimensional Gaussian binary classification | suppressor-feature importance: 0 | Helped |
| Use pixel flipping to measure importance | Bayes optimal linear classifier · Two-dimensional Gaussian binary classification | suppressor-feature importance: \alpha^2 c^2 s_1^2 | Hurt |
| Use permutation feature importance | Bayes optimal linear classifier · Two-dimensional Gaussian binary classification | suppressor-feature importance: 2 \alpha^2 c^2 s_1^2 | Hurt |
| Use partial dependency plots | Bayes optimal linear classifier · Two-dimensional Gaussian binary classification | suppressor-feature importance: -\alpha c s_1 s_{2}^{-1}x_2 | Hurt |
| Use M-plots based on conditional expectations | Bayes optimal linear classifier · Two-dimensional Gaussian binary classification | suppressor-feature importance: 0 | Helped |
| Calculate Shapley values using R^2 with subset models | Bayes optimal linear classifier · Two-dimensional Gaussian binary classification | suppressor-feature importance: \frac{\alpha - 1}{2(s_1^2 + 1)^{1/2}} | Hurt |
| Calculate Shapley values using R^2 with the full model only | Bayes optimal linear classifier · Two-dimensional Gaussian binary classification | suppressor-feature importance: 0 | Helped |
| Use SHAP with marginal expectations | Bayes optimal linear classifier · Two-dimensional Gaussian binary classification | suppressor-feature importance: -\alpha c s_1 s_2^{-1} x_2 | Hurt |
| Use SHAP with conditional expectations | Bayes optimal linear classifier · Two-dimensional Gaussian binary classification | suppressor-feature importance: \frac{\alpha c^2}{2} h(x_1) - \frac{\alpha c s_1}{2 s_2} x_2 | Hurt |
| Use FIRM, which incorporates training-data statistics | Bayes optimal linear classifier · Two-dimensional Gaussian binary classification | suppressor-feature importance: 0 | Helped |
| Use integrated gradients | Bayes optimal linear classifier · Two-dimensional Gaussian binary classification | suppressor-feature importance: - \frac{\alpha c s_1}{2 s_2} (x_2^2 - (\mathbf{x}')^2) | Hurt |
Closest papers by meaning
- Navigating the Maze of Explainable AI: A Systematic Approach to Evaluating Methods and Metrics (NeurIPS 2024 · similarity 0.84)
- F-Fidelity: A Robust Framework for Faithfulness Evaluation of Explainable AI (ICLR 2025 · similarity 0.84)
- Evaluate with the Inverse: Efficient Approximation of Latent Explanation Quality Distribution (AAAI 2025 · similarity 0.83)
- Probabilistic Explanations for Linear Models (AAAI 2025 · similarity 0.83)
- Local vs. Global Interpretability: A Computational Complexity Perspective (ICML 2024 · similarity 0.83)
- Minimizing False-Positive Attributions in Explanations of Non-Linear Models (NeurIPS 2025 · similarity 0.83)
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.