阅读：General Quantification of Covariate and Concept Shifts

• 模式: 沉浸阅读高引用论文 (roll=41)
• 时间: 2026-09-11 20:30:10

阅读记录：General Quantification of Covariate and Concept Shifts

• 来源: arxiv | 年份: N/A | 引用: 0 | 链接: http://arxiv.org/abs/2609.11918v1

论文摘要（原始）

Generalization under distribution shift remains a core challenge in modern machine learning, yet existing learning bound theory is limited to narrow, idealized settings and is non-estimable from samples. In this paper, we bridge the gap between theory and practical applications. We first show that existing definition of concept shift breaks when the source and target supports mismatch. Leveraging entropic optimal transport, we propose a key notion: $γ^{}\!$-concept shifts, and derive a general error bound unifying covariate and $γ^{}\!$-concept shifts, which applies to broad loss functions, label spaces, and stochastic labeling. We further develop estimators for these shifts with concentration guarantees, and the DataShifts algorithm, which can quantify distribution shifts and estimate the error bound in most applications - a rigorous and general tool for analyzing learning error under distribution shift.

AI 概括

Lyco 推眼镜，清清嗓子，开始唠嗑：

---

1️⃣ 背景：分布偏移是个“老大难”，理论却一直在“象牙塔”里打转

大家都知道，模型上线后遇到的数据分布（协变量分布 $P(X)$、标签分布 $P(Y|X)$）跟训练时不一样，性能就会跳水。现有的学习界理论要么只盯着 covariate shift（协变量偏移），要么只盯着 concept shift（概念偏移），还得假设源域和目标域的支撑集完全重合——这在现实里基本是童话。更坑的是，那些界里的关键量（比如密度比、全变差）根本没法从有限样本里估出来，理论指导不了实践，实践也验证不了理论。

---

2️⃣ 核心洞见：传统“概念偏移”定义在支撑集不对齐时直接GG

作者第一件事是找茬：把经典定义 $P_S(Y|X) \neq P_T(Y|X)$ 拿到显微镜下一看，只要源域和目标域的 $X$ 支撑集不完全重合（比如目标域冒出了训练集从没见过的特征），条件概率根本没法比，定义直接失效。这解释为啥以前的界在“开放世界”场景下全瘫痪。

---

3️⃣ 方法：把熵正则化最优传输（Entropic OT） 拉来当“万能胶水”

• γ\-concept shift：用 OT 耦合 $\pi^$ 把源域和目标域的联合分布 $P(X,Y)$ “对齐”起来，哪怕支撑集不重合也能算出一个可比、可估的偏移量。
• 统一误差界：一个公式同时塞进 covariate shift、γ\*-concept shift、任意损失函数、离散/连续标签、甚至随机标签噪声。界的形式长得像

$$

\mathcal{E}_T(h) \le \mathcal