Novel Class

Novel class discovery (NCD) focuses on identifying and classifying previously unseen data categories within machine learning models, leveraging existing knowledge of known classes to improve performance. Current research emphasizes techniques like prototype-based methods, spectral analysis, and contrastive learning to address challenges such as cross-domain data distributions, catastrophic forgetting, and limited labeled data for novel classes. These advancements are crucial for improving the adaptability and robustness of machine learning systems in dynamic environments, with applications ranging from remote sensing and land cover mapping to object detection and natural language processing.