General Quantification of Covariate and Concept Shifts
协变量偏移与概念偏移的通用量化
机构 * South China University of Technology(华南理工大学)
AI总结 本文提出基于熵正则化最优传输的$\gamma^{*}$-概念偏移定义,统一协变量与概念偏移的误差界,并开发可估计的DataShifts算法,为分布偏移下的学习误差提供通用量化工具。
Comments 38 pages, 9 figures, accepted at the 43rd International Conference on Machine Learning (ICML 2026)