Research map / Learning theory and algorithms / Prediction and decision losses
Which Explanation Should I Choose? A Function Approximation Perspective to Characterizing Post Hoc Explanations
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 additive continuous noise with SmoothGrad or Vanilla Gradients | Linear models on continuous inputs · WHO | model recovery | Helped |
| Use additive continuous noise with SmoothGrad or Vanilla Gradients | Logistic regression on continuous inputs · HELOC | model recovery | Helped |
| Use multiplicative continuous noise with Integrated Gradients or Gradient x Input | Linear models on continuous inputs · WHO | model recovery | Hurt |
| Use multiplicative binary noise with LIME, KernelSHAP, or Occlusion | Linear models on continuous inputs · WHO | model recovery | Hurt |
| Evaluate explanations with binary perturbations | LIME, KernelSHAP, Occlusion, Integrated Gradients, and Gradient x Input · WHO, neural network with three hidden layers | bottom-k perturbation performance | Helped |
| Evaluate explanations with continuous perturbations | SmoothGrad and Vanilla Gradients · WHO, neural network with three hidden layers | bottom-k perturbation performance | Helped |
| Match the explanation method's neighborhood to the evaluation perturbations | Post hoc explanation methods · WHO and HELOC datasets | perturbation-test performance | Mixed |
| Modify the gradient-matching loss to compare gradients at the perturbed input rather than at the noise variable (ablation) | Integrated Gradients and Gradient x Input · Continuous inputs | model recovery | Helped |
| Modify the binary-noise loss to compare f and g at the perturbed input rather than comparing g to the noise variable (ablation) | Binary-noise explanation methods · Continuous inputs, linear models | model recovery | Helped |
| Use binary noise with the modified loss for periodic functions with large frequency components (ablation) | Binary-noise explanation methods · Continuous inputs | model recovery | Hurt |
Closest papers by meaning
- Axiomatic Aggregations of Abductive Explanations (AAAI 2024 · similarity 0.88)
- Even-if Explanations: Formal Foundations, Priorities and Complexity (AAAI 2025 · similarity 0.87)
- Local vs. Global Interpretability: A Computational Complexity Perspective (ICML 2024 · similarity 0.86)
- Dissenting Explanations: Leveraging Disagreement to Reduce Model Overreliance (AAAI 2024 · similarity 0.86)
- BEE: Metric-Adapted Explanations via Baseline Exploration-Exploitation (AAAI 2025 · similarity 0.86)
- Explaining Probabilistic Models with Distributional Values (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.