# Selective Prediction and Abstention

2 related articles

Subscribe to updates

arXiv Oct 26, 2026
95% intervals calibrated on random splits fail outside the training envelope: hard-shift coverage falls to 0.40–0.75 across seven engineering datasets
Prediction intervals for engineering regression are usually calibrated on random splits, while deployment happens outside training conditions. A benchmark splits seven public engineering datasets into in-distribution, soft-shift and hard-shift rungs along physically meaningful shift variables, and compares 12 uncertainty quantification methods from five families tuned on source data only. In distribution, conformal methods cover at least 0.946 on every dataset; under hard shift, mean coverage falls to 0.40–0.75 for the 11 methods that always return an interval, and no systematic difference in method ranks is detected.
Prediction intervals for engineering regression are usually calibrated on random splits, while deployment happens outside training conditions. A benchmark splits seven public engineering datasets into in-distribution, soft-shift and hard-shift rungs along physically meaningful shift variables, and compares 12 uncertainty quantification methods from five families tuned on source data only. In distribution, conformal methods cover at least 0.946 on every dataset; under hard shift, mean coverage falls to 0.40–0.75 for the 11 methods that always return an interval, and no systematic difference in method ranks is detected.
De Montfort University
Ahmad Lawal（First author）
Uncertainty Quantification Conformal Prediction Weighted Conformal Prediction Distribution Shift Selective Prediction and Abstention
researchsquare.com Oct 21, 2026
Entanglement diagnosis stops guessing one label: finite-shot Pauli records yield calibrated verdicts up to 20 qubits
Multipartite-entanglement diagnosis now takes finite-shot Pauli records directly as input instead of starting from full tomography. Earlier learning-based methods tended to return a single label even when the evidence was thin. This protocol combines binomial Pauli estimators, task-informed observable designs, class-conditional conformal prediction and out-of-distribution abstention to output a single label, a prediction set or an abstention; on synthetic benchmarks, phase-aware global probes reached 0.968 balanced accuracy for perturbed-GHZ discrimination at 20 qubits.
Multipartite-entanglement diagnosis now takes finite-shot Pauli records directly as input instead of starting from full tomography. Earlier learning-based methods tended to return a single label even when the evidence was thin. This protocol combines binomial Pauli estimators, task-informed observable designs, class-conditional conformal prediction and out-of-distribution abstention to output a single label, a prediction set or an abstention; on synthetic benchmarks, phase-aware global probes reached 0.968 balanced accuracy for perturbed-GHZ discrimination at 20 qubits.
Shenyang University of Technology Yanbian University
Wenlong Sun（First author）
Finite-Shot Pauli Measurement Model Calibration Multipartite Entanglement Diagnosis Selective Prediction and Abstention Conformal Prediction

You’re all caught up

95% intervals calibrated on random splits fail outside the training envelope: hard-shift coverage falls to 0.40–0.75 across seven engineering datasets

Entanglement diagnosis stops guessing one label: finite-shot Pauli records yield calibrated verdicts up to 20 qubits