Bulk-Calibrated Credal Ambiguity Sets: Fast, Tractable Decision Making under Out-of-Sample Contamination
Researchers have developed a new framework called bulk-calibrated credal ambiguity sets to improve decision-making under out-of-sample contamination in distributionally robust optimization (DRO). This method learns a high-mass bulk set from data while separately bounding contamination in the tail, leading to a closed-form, finite objective that can be optimized using tractable linear or second-order cone programs. Experiments on various tasks like inventory control and text classification demonstrated competitive robustness-accuracy trade-offs and efficient optimization times. AI