DataShifts Algorithm Quantifies Covariate and Concept Shifts
September 11, 2026
The DataShifts algorithm uses entropic optimal transport to unify covariate and concept shifts under a general error bound. This provides a way to estimate learning error under distribution shift in practical applications where source and target supports mismatch.
HOW THIS AFFECTS YOU
●
builderYou can use this algorithm to quantify how much your model's performance will degrade under distribution shift.
●
researcherThe proposed $\gamma^{*}$-concept shift notion provides a more robust theoretical framework for non-idealized settings.