<ul><li>Machine learning models often experience performance decay when deployed in new contexts, with some subgroups being more affected than others.</li><li>Understanding the reasons behind large performance differences in subgroups is essential for implementing corrective measures efficiently.</li><li>A new framework, Subgroup-scanning Hierarchical Inference Framework for performance drifT (SHIFT), aims to identify groups with significant performance decay and explain the causes behind it.</li><li>Real-world experiments show that SHIFT helps in pinpointing interpretative subgroups affected by performance decay and proposing effective mitigation strategies.</li></ul>

"Who experiences large model decay and why?" A Hierarchical Framework for Diagnosing Heterogeneous Performance Drift

Discover more