Identifying when a model receives input data that differs significantly from its training distribution.