Research map / Learning theory and algorithms / Prediction and decision losses
Predictive Multiplicity in Probabilistic Classification
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 |
|---|---|---|---|
| Decrease separability | Logistic regression risk prediction on synthetic data · Synthetic data, varying separability | ambiguity: 1% to 21% | Helped |
| Place outliers closer to the discriminant boundary | Logistic regression risk prediction on synthetic data · Synthetic data, outliers and margin distance | ambiguity | Helped |
| Introduce a majority-minority group structure | Logistic regression risk prediction on synthetic data · Synthetic data, majority-minority structure | ambiguity | Mixed |
| Use near-optimal models defined by training loss rather than AUC | Probabilistic classification tasks · Real-world risk assessment datasets | ambiguity | Mixed |
| Use near-optimal models defined by training loss rather than AUC | Probabilistic classification tasks · Real-world risk assessment datasets | discrepancy | Mixed |
| Focus on examples with more than 20 duplicates rather than unique examples | Probabilistic classification tasks · Real-world risk assessment datasets | ambiguity: less than 10% | Hurt |
| Evaluate risk predictions for Hispanic individuals rather than African American or Caucasian individuals | COMPAS arrest risk prediction · COMPAS arrest, ethnic subgroups | ambiguity: 39% for African Americans and 49% for Caucasians, compared to 98% for Hispanics | Mixed |
| Allow variation across near-optimal models | Logistic regression risk prediction · Real-world risk assessment datasets | viable prediction range | Helped |
Closest papers by meaning
- Dropout-Based Rashomon Set Exploration for Efficient Predictive Multiplicity Estimation (ICLR 2024 · similarity 0.86)
- Reconciling Model Multiplicity for Downstream Decision Making (ICLR 2025 · similarity 0.85)
- Quantifying Prediction Consistency Under Fine-tuning Multiplicity in Tabular LLMs (ICML 2025 · similarity 0.84)
- Imputation for prediction: beware of diminishing returns. (ICLR 2025 · similarity 0.84)
- Random features models: a way to study the success of naive imputation (ICML 2024 · similarity 0.83)
- Regretful Decisions under Label Noise (ICLR 2025 · similarity 0.82)
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.