Research map / Learning theory and algorithms / Prediction and decision losses
Implications of Model Indeterminacy for Explanations of Automated Decisions
Marc-Etienne Brunet, Ashton Anderson, Richard Zemel
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 |
|---|---|---|---|
| source of indeterminacy: Rashomon set instead of underspecification | Dinput Shapley explanations · UCI Credit Card | top-5 CDC (%): 9.3 -> 54.9 | Hurt |
| source of indeterminacy: Rashomon set instead of underspecification | Dinput Shapley explanations · UCI Credit Card | top-3 SSD (%): 51.1 -> 79.4 | Hurt |
| source of indeterminacy: Rashomon set instead of underspecification | Dinput Shapley explanations · Porto Seguro Safe Driver | top-5 CDC (%): 0.6 -> 23.6 | Hurt |
| explain top-5 instead of top-3 features | Dinput Shapley, underspecification · UCI Credit Card | SSD (%): 51.1 -> 67.2 | Hurt |
| explain top-5 instead of top-3 features | Dinput Shapley, Rashomon · Give Me Some Credit | CDC (%): 11.1 -> 29.0 | Hurt |
| Dmean Shapley instead of Dinput Shapley | Rashomon sets · UCI Credit Card | top-3 CDC (%): 36.5 -> 46.6 | Hurt |
| Dmean Shapley instead of Dinput Shapley | underspecification sets · UCI Credit Card | top-3 SSD (%): 51.1 -> 72.1 | Hurt |
| two approximately equivalent models at 50% acceptance threshold | classifiers · three credit datasets | decision disagreement (%): 4.35-12.7 | Hurt |
| use epistemic uncertainty of the prediction to predict explanation consistency | approximately equivalent models · UCI Credit Card and Safe Driver | Spearman correlation with SA: -0.17 to 0.23 | No clear effect |
| use epistemic uncertainty of the prediction to predict explanation consistency | approximately equivalent models · Give Me Some Credit | Spearman correlation with SA: -0.34 to -0.58 | Mixed |
Closest papers by meaning
- Dissenting Explanations: Leveraging Disagreement to Reduce Model Overreliance (AAAI 2024 · similarity 0.89)
- Reliable Post hoc Explanations: Modeling Uncertainty in Explainability (NeurIPS 2021 · similarity 0.87)
- Reconciling Model Multiplicity for Downstream Decision Making (ICLR 2025 · similarity 0.86)
- Local vs. Global Interpretability: A Computational Complexity Perspective (ICML 2024 · similarity 0.84)
- Causality can systematically address the monsters under the bench(marks) (NeurIPS 2026 · similarity 0.84)
- Rethinking Aleatoric and Epistemic Uncertainty (ICML 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.
