[论文] The cost of enforcing long-run coverage: a phase transition in online ...
论文概要
研究领域: ML
作者: Yu Gui, Rina Foygel Barber, Cong Ma
发布时间: 2026-11-16
arXiv: 2611.21542
中文摘要
测试时增强(TTA)在推理期间聚合多个变换视图的模型预测。由于组织病理学图像缺乏规范方向,D4 TTA——对正方形的 8 种旋转和反射进行池化——是计算病理学中的标准默认做法。然而,具有内在旋转不变性的自监督组织病理学基础模型(FM)的兴起,使其持续效用受到质疑。我们在四个 patch 级和两个全切片图像(WSI)数据集上对 9 个组织病理学 FM 和 6 个通用模型进行了基准测试。我们发现 D4 TTA 仍然是可靠的后期改进,但收益从通用骨干的 +1.98 个百分点降至组织病理学 FM 在 TCGA-UT 上的 +0.50 个百分点,且在最大模型上进一步缩小。为消除计算开销,我们的实验表明熵路由——仅对不确定的预测选择性应用 TTA——恢复了大部分收益,同时仅增强约 13% 的样本。最后,我们表明 patch 级 TTA 的改进不会迁移到切片级的多实例学习(MIL)。
原文摘要
Consider an online prediction setting, where at each time, we are given a new feature vector and asked to provide a prediction interval or prediction set that will likely contain the corresponding response. In this setting, if the data points are i.i.d. (or more generally, exchangeable), conformal prediction provides a mechanism for building valid prediction sets around the output of any trained model, offering coverage probability \(\geq 1-α\) at each time \(t\). In practice, however, issues such as distribution drift, changepoints, or other challenges may cause violations of the i.i.d. assumption. In this adversarial setting, online conformal prediction methods can still offer guarantees on long-run empirical coverage. However, there is a trade-off between these two types of goals: enforcing...
自动采集于 2026-11-18
#论文 #arXiv #ML #小凯