Research map / Probabilistic and scientific modeling / Causal discovery
Asymmetric Shapley values: incorporating causal knowledge into model-agnostic explainability
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 on-manifold instead of off-manifold marginalization | Shapley explanations · Census Income | feature attribution consistency with data correlations | Helped |
| Order known causal ancestors before descendants in the attribution permutations | Asymmetric Shapley values · Census Income | attribution to sex: roughly 3% | Helped |
| Order department before gender when attributing importance | Asymmetric Shapley values · Synthetic college admissions fair data | gender attribution after accounting for department | Helped |
| Order department before gender when attributing importance | Asymmetric Shapley values · Synthetic college admissions unfair data | detection of unresolved gender discrimination | Helped |
| Impose time order on feature permutations | Asymmetric Shapley values · Epileptic Seizure Recognition | sparsity of feature attribution: roughly 10^2 after 25 (out of 178) steps | Helped |
| Use cumulative time-ordered ASVs to estimate accuracy attributable to observed time steps | Asymmetric Shapley values · Epileptic Seizure Recognition | agreement with empirical accuracies | Helped |
| Vary LSTM dimension, batch size, and patience up or down by a factor of 2 (ablation) | On-manifold ASVs for the seizure model · Epileptic Seizure Recognition | ASV results: by a factor of 2 | No clear effect |
Closest papers by meaning
- Practical do-Shapley Explanations with Estimand-Agnostic Causal Inference (NeurIPS 2025 · similarity 0.88)
- Incorporating Information into Shapley Values: Reweighting via a Maximum Entropy Approach (ICML 2024 · similarity 0.88)
- Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable? (ICLR 2025 · similarity 0.88)
- A New Approach to Backtracking Counterfactual Explanations: A Unified Causal Framework for Efficient Model Interpretability (ICML 2025 · similarity 0.87)
- Local vs. Global Interpretability: A Computational Complexity Perspective (ICML 2024 · similarity 0.87)
- Gnothi Seauton: Empowering Faithful Self-Interpretability in Black-Box Transformers (ICLR 2025 · similarity 0.86)
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.