Class Hierarchy

Class hierarchies represent the structured relationships between categories, a crucial aspect in many machine learning tasks. Current research focuses on integrating hierarchical information into model training to improve performance and interpretability, employing techniques like optimal transport for embedding hierarchies and graph neural networks for representing hierarchical relationships. This work is significant because effectively leveraging class hierarchies enhances the accuracy and explainability of models across diverse applications, from image recognition and natural language processing to robotics and continual learning. The development of efficient algorithms and robust model architectures for handling hierarchical data is a key area of ongoing investigation.