Research map / Learning theory and algorithms / Prediction and decision losses
Bulk-Calibrated Credal Ambiguity Sets: Fast, Tractable Decision Making under Out-of-Sample Contamination
Mengqi Chen, Thomas Berrett, Theodoros Damoulas, Michele Caprio
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 LV bulk-calibrated contamination objective instead of Wasserstein DRO | linear regression with absolute loss · California housing East->West shift | MAE (x10^4): 12.3 -> 10.7 | Improved |
| use LV objective instead of Wasserstein DRO | linear regression with absolute loss · California housing East->West shift | CVaR2% (x10^4): 26.5 -> 23.9 | Improved |
| add LV contamination layer on top of GroupDRO (LV-Group) | GroupDRO linear head · CivilComments | worst-group accuracy: 0.456 -> 0.516 | Improved |
| add LV contamination layer on top of GroupDRO (LV-Group) | GroupDRO linear head · CivilComments | mean accuracy: 0.770 -> 0.828 | Improved |
| train GroupDRO on bulk-filtered data only (without the LV contamination term) (ablation) | GroupDRO · CivilComments | worst-group accuracy: 0.456 -> 0.453 | No clear effect |
| use LV-Group instead of ERM | linear head · CivilComments | worst-group accuracy: 0.022 -> 0.516 (mean acc 0.891 -> 0.828) | Improved |
| use LV objective with fewer SAA samples (truncated 0.5M) than KL baselines | newsvendor Bayesian centre · Student-t newsvendor | total run time (s): 3.6 -> 1.3 vs KL-BASPP | Improved |
| use LV under no test contamination with large epsilon_LV | LV newsvendor · Student-t newsvendor (0% contamination) | MSD | Hurt |
| use LV under contaminated test settings (10%, 20%) | newsvendor · Student-t newsvendor (10% and 20% contamination) | MSD | Improved |
| increase KL tolerance beyond finite-scenario limit in KL-BDRO (ablation) | KL-BDRO · Student-t newsvendor | OOS mean-variance | No clear effect |
| vary tail budget gamma around minimum certifiable value (ablation) | LV newsvendor · Student-t newsvendor | OOS performance | No clear effect |
Closest papers by meaning
- Decision Making under the Exponential Family: Distributionally Robust Optimisation with Bayesian Ambiguity Sets (ICML 2025 · similarity 0.88)
- Bayesian Nonparametrics Meets Data-Driven Distributionally Robust Optimization (NeurIPS 2024 · similarity 0.84)
- Credal Prediction based on Relative Likelihood (NeurIPS 2025 · similarity 0.84)
- Distributionally Robust Performative Optimization (NeurIPS 2025 · similarity 0.84)
- Designing Ambiguity Sets for Distributionally Robust Optimization Using Structural Causal Optimal Transport (AAAI 2025 · similarity 0.84)
- Distributionally Robust Optimization with Bias and Variance Reduction (ICLR 2024 · similarity 0.84)
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.
