New Category

Research on novel category discovery focuses on developing methods to identify and classify new categories within unlabeled data, leveraging knowledge from existing labeled data. Current efforts concentrate on addressing challenges like imbalanced datasets, limited labeled examples, and the need for robust handling of both known and unknown categories, employing techniques such as self-training, optimal transport, and multimodal learning with models like CLIP. This work is significant for advancing machine learning capabilities in areas like object recognition, document classification, and product categorization, particularly in scenarios with limited or incomplete data.