让置信度改变，而非预测改变：面向事后校准的保预测修复方法

摘要

事后校准可以修正报告置信度，但多类校准器也可能改变相应的前1预测。准确率仅捕捉这些改变对正确性的净效应，而非预测发生变更的频率；前1预测变化率（Top-1 Prediction Change Rate, TPCR）则专门度量该频率。我们提出“用于前1决策保持的校准器输出修复”（Calibrator-Output Repair for Top-1 Decision Preservation, CORD），这是首个通过修复完整校准概率向量来强制实现精确预测保持的拟合后适配器。仅依据原始输出与校准输出，CORD即可确定原前1类别所分配的概率质量。校准后的条件分布将剩余质量分配至其他类别，由此得到的修复向量，其自身的argmax可恢复原始预测。在校准划分上，CORD协调修复后的概率质量，以在可达到的情况下保留校准确输出对原始预测的平均质量。该适配器既不改变已拟合的校准器，也不改变其直接输出，不拟合额外的监督映射，且无需用户或验证集调节的超参数。在CIFAR-10/100与ImageNet-1K上，CORD通过构造实现零TPCR，并在每个数据集中相对对应直接输出降低平均ECE、NLL与布里尔分数；在分布偏移和不同校准集规模下，配对增益依然持续存在。因此，CORD将保持约束从校准器拟合中移除，并把原始决策的精确恢复指派给随后的输出修复。我们的代码可在 https://github.com/labhai/CORD 获取。

English

Post-hoc calibration corrects reported confidence, yet a multiclass calibrator can also change the associated top-1 prediction. Accuracy captures only the net effect of these changes on correctness, not how often predictions change; the Top-1 Prediction Change Rate (TPCR) instead measures this frequency. We propose Calibrator-Output Repair for Top-1 Decision Preservation (CORD), the first post-fit adapter to impose exact prediction preservation by repairing the full calibrated probability vector. From the original and calibrated outputs alone, CORD determines the mass assigned to the original top-1. The calibrated conditional distribution allocates the remaining mass over the other classes, yielding a repaired vector whose own argmax recovers the original prediction. On the calibration split, CORD coordinates the repaired masses to retain the calibrated outputs' mean mass on original predictions whenever attainable. The adapter alters neither the fitted calibrator nor its direct output, fits no additional supervised map, and requires no user- or validation-tuned hyperparameter. Across CIFAR-10/100 and ImageNet-1K, CORD attains zero TPCR by construction and lowers mean ECE, NLL, and Brier relative to the corresponding direct outputs in every dataset; paired gains persist under distribution shift and across calibration-set sizes. CORD thus removes the preservation constraint from calibrator fitting and assigns exact recovery of the original decision to subsequent output repair. Our code is available at https://github.com/labhai/CORD.