Improving machine learning accuracy when data changes unexpectedly
General Quantification of Covariate and Concept Shifts
Machine LearningArtificial Intelligence
Summary
Machine learning models often struggle when the data they were trained on looks different from what they see later, a situation called distribution shift. Existing theories don't fully explain how errors happen in realistic cases, especially when the types of data don't match perfectly. The authors introduce a new way to measure changes in data and labels using a mathematical tool called entropic optimal transport. This leads to better predictions of a model's error when facing different types of shifts in data. They also created a practical method, DataShifts, to estimate these changes and error bounds from real data.
distribution shiftmachine learninggeneralization errorcovariate shiftconcept shiftentropic optimal transporterror boundDataShifts algorithmstochastic labelinglabel space
Authors
Hongbo Chen, Li Charlie Xia
Abstract
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.