Decoupled Conformal Optimisation: Efficient Prediction Sets via Independent Tuning and Calibration
Two new research papers introduce novel approaches to conformal prediction, a method for quantifying uncertainty in machine learning models. The first paper, "Decoupled Conformal Optimisation," proposes a train-tune-calibrate framework that uses independent data splits for structural selection and final calibration, leading to smaller prediction sets and interval widths on various benchmarks. The second paper, "Decomposition-Based Modular Conformal Prediction," extends conformal prediction to two-stage modeling, allowing for the attribution of uncertainty to specific pipeline stages and offering diagnostic advantages over standard methods. AI
IMPACT These new conformal prediction techniques offer improved uncertainty quantification and diagnostic capabilities for machine learning models.