Research map / Learning theory and algorithms / Prediction and decision losses
TRIAGE: Characterizing and auditing training data for improved regression
Nabeel Seedat, Jonathan Crabbé, Zhaozhi Qian, Mihaela van der Schaar
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 |
|---|---|---|---|
| train on well-estimated samples only instead of OE+UE samples | TRIAGE sculpting · Prostate SEER->CUTRACT, |Dcal|=10 | test MSE: 0.088 -> 0.051 | Improved |
| sculpt Dtrain with TRIAGE instead of using full Dtrain | regressor · Prostate SEER->CUTRACT, |Dcal|=10 | test MSE: 0.064 -> 0.051 | Improved |
| TRIAGE sculpting instead of NGBoost uncertainty sculpting | regressor · Prostate, |Dcal|=100 | test MSE: 0.196 -> 0.046 | Improved |
| TRIAGE sculpting instead of Gaussian process sculpting | regressor · Prostate, |Dcal|=500 | test MSE: 0.089 -> 0.046 | Improved |
| TRIAGE sculpting instead of error-based sculpting | regressor · Prostate, |Dcal|=10 | test MSE: 0.066 -> 0.051 | Improved |
| TRIAGE sculpting instead of combining Dtrain and Dcal | regressor · Prostate, |Dcal|=50 | test MSE: 0.055 -> 0.046 | Improved |
| score with final-checkpoint residuals only (no training dynamics) (ablation) | TRIAGE · 10 regression datasets | Spearman consistency: 0.91 | No clear effect |
| use TRIAGE scores instead of VoG, GradN, Loss | scoring method · 10 regression datasets | Spearman rank correlation: 0.91 | Improved |
| add TRIAGE as fine-grained filter keeping well-estimated samples on top of baseline subsets | Residuals, Loss, VoG, GradN filters · regression datasets | test MSE | Improved |
| select synthetic dataset with higher TRIAGE retention (TVAE over CTGAN) | TSTR regressor · Concrete | MAE: 0.19 -> 0.11 | Improved |
| select synthetic dataset with higher TRIAGE retention (TVAE over CTGAN) | TSTR regressor · Bike | MAE: 0.13 -> 0.09 | Improved |
| select synthetic dataset with higher TRIAGE retention (TVAE over CTGAN) | TSTR regressor · MIMIC | MAE: 0.05 -> 0.08 | Hurt |
Closest papers by meaning
- quanda: An Interpretability Toolkit for Training Data Attribution Evaluation (NeurIPS 2026 · similarity 0.83)
- Final-Model-Only Data Attribution with a Unifying View of Gradient-Based Methods (NeurIPS 2025 · similarity 0.82)
- Reproduce, Replicate, Reevaluate. The Long but Safe Way to Extend Machine Learning Methods (AAAI 2024 · similarity 0.82)
- A Unified Comparative Study with Generalized Conformity Scores for Multi-Output Conformal Regression (ICML 2025 · similarity 0.82)
- Robust Conformal Outlier Detection under Contaminated Reference Data (ICML 2025 · similarity 0.82)
- Embarrassingly Simple Dataset Distillation (ICLR 2024 · 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.
