Hierarchical Classification

Hierarchical classification studies how models can use label taxonomies to make predictions that are useful at both coarse and fine levels. Common approaches add hierarchy-aware losses, separate prediction heads, graph or geometric label representations, and coarse-to-fine decision procedures to standard classifiers. These methods are used in settings such as medical diagnosis, image and text recognition, and large code or species taxonomies, where errors between nearby classes may be less serious than errors across broad groups. Open challenges include preventing early mistakes from cascading, handling incomplete or shifting labels, and balancing fine-grained accuracy with consistency, calibration, and safe fallback to broader categories.