General Quantification of Covariate and Concept Shifts
The paper proposes an estimable framework for measuring distribution shift when source and target data do not cleanly overlap.
Chen and Xia argue that the usual definition of concept shift fails when source and target supports mismatch. They introduce gamma*-concept shifts using entropic optimal transport, then derive an error bound that combines covariate shift and that concept-shift measure. The work also gives sample-based estimators with concentration guarantees and a DataShifts algorithm for estimating the shift measures and bound. ArXiv · AI/CL/LG's note
Chen and Xia argue that the usual definition of concept shift fails when source and target supports mismatch. They introduce gamma*-concept shifts using entropic optimal transport, then derive an error bound that combines covariate shift and that concept-shift measure. The work also gives sample-based estimators with concentration guarantees and a DataShifts algorithm for estimating the shift measures and bound. ArXiv · AI/CL/LG's note
score 4