In decision-making guided by machine learning, decision-makers may take identical actions in contexts with identical predicted outcomes. Conformal prediction helps decision-makers quantify uncertainty in point predictions of outcomes, allowing for better risk management for actions. Motivated by this perspective, we introduce \textit{Self-Consistent Conformal Prediction} for regression, which combines two post-hoc approaches -- Venn-Abers calibration and conformal prediction -- to provide calibrated point predictions and compatible prediction intervals that are valid conditional on model predictions. Our procedure can be applied post-hoc to any black-box model to provide predictions and inferences with finite-sample prediction-conditional guarantees. Numerical experiments show our approach strikes a balance between interval efficiency and conditional validity.

翻译：在基于机器学习的决策过程中，面对预测结果相同的上下文，决策者可能采取相同的行动。共形预测帮助决策者量化对结果点预测中的不确定性，从而优化行动的风险管理。基于这一视角，我们针对回归问题提出了\textit{自洽共形预测}方法，该方法融合了两种后处理策略——Venn-Abers校准与共形预测——以提供校准后的点预测以及兼容的预测区间，这些区间在模型预测条件下具有有效性。我们的流程可后验应用于任意黑箱模型，在有限样本下提供具有预测条件保障的预测与推断。数值实验表明，本方法在区间效率与条件有效性之间取得了良好平衡。