Research map / Learning theory and algorithms / Prediction and decision losses
Predicting Rare Events by Shrinking Towards Proportional Odds
Gregory Faletto, Jacob Bien
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 |
|---|---|---|---|
| Use PRESTO, which shares strength across adjacent decision boundaries while allowing sparse coefficient differences | Rare-event probability estimation · PreDiabetes, age cutoffs 30–65 | estimated rare class MSE: At age cutoffs 30, 35, 40, 45, 50, 55, 60, and 65: PRESTO 0.000943, 0.005013, 0.017307, 0.060896, 0.124000, 0.234906, 0. | Improved |
| Use PRESTO rather than logistic regression on the rare category | Rare-event probability estimation · Soup tasting | estimated rare probability MSE: 0.0157 -> 0.0096 | Improved |
| Use PRESTO rather than the proportional odds model | Rare-event probability estimation · Soup tasting | estimated rare probability MSE: 0.0135 -> 0.0096 | Improved |
| Use PRESTO rather than logistic regression on the rare category | Rare-event probability estimation · Sparse differences synthetic, rare class proportion 0.71%, sparsity 1/2 | empirical rare probability MSE: 8.89e-05 (2.5e-06) -> 5.22e-05 (1.9e-06) | Improved |
| Use PRESTO rather than proportional odds | Rare-event probability estimation · Sparse differences synthetic, rare class proportion 0.71%, sparsity 1/2 | empirical rare probability MSE: 7.50e-05 (3e-06) -> 5.22e-05 (1.9e-06) | Improved |
| Use PRESTO rather than logistic regression on the rare category | Rare-event probability estimation · Uniform differences synthetic, rare class proportion 0.62% | empirical rare probability MSE: 6.51e-05 (1.7e-06) -> 2.86e-05 (8.3e-07) | Improved |
| Use PRESTO rather than proportional odds | Rare-event probability estimation · Uniform differences synthetic, rare class proportion 0.62% | empirical rare probability MSE: 3.43e-05 (9.8e-07) -> 2.86e-05 (8.3e-07) | Improved |
| Replace the L1 fusion penalty with an L2 ridge penalty (PRESTO_L2) (ablation) | PRESTO · Sparse differences synthetic, rare class proportion 1%, sparsity 1/3 | empirical rare probability MSE: 6.05e-05 (2.1e-06) -> 6.25e-05 (2.4e-06) | Hurt |
| Replace the L1 fusion penalty with an L2 ridge penalty (PRESTO_L2) (ablation) | PRESTO · Sparse differences synthetic, rare class proportion 1.17%, sparsity 1/2 | empirical rare probability MSE: 9.87e-05 (2.9e-06) -> 1.17e-04 (4.2e-06) | Hurt |
Closest papers by meaning
- SimPro: A Simple Probabilistic Framework Towards Realistic Long-Tailed Semi-Supervised Learning (ICML 2024 · similarity 0.84)
- Nearly Optimal Sample Complexity for Learning with Label Proportions (ICML 2025 · similarity 0.84)
- SLACE: A Monotone and Balance-Sensitive Loss Function for Ordinal Regression (AAAI 2025 · similarity 0.83)
- Rethinking Learning from Label Proportions via Moment Matching (NeurIPS 2026 · similarity 0.83)
- Improving Model Probability Calibration by Integration of Large Data Sources with Biased Labels (AAAI 2025 · similarity 0.83)
- Treeffuser: probabilistic prediction via conditional diffusions with gradient-boosted trees (NeurIPS 2024 · 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.
