Beyond the Performance Illusion: Structure-Aware Stratified Partitioning and Curriculum Distributionally Robust Optimization for Spatially Correlated Domains

Researchers have identified that standard AI evaluation methods often fail in spatiotemporally correlated environments, such as aerial surveillance, because they incorrectly assume data points are independent. To address this, the study introduces new partitioning and optimization techniques that account for the underlying spatial structure of the data to ensure more accurate performance metrics.