Recent research advances in Novel Class Discovery

Novel Class Discovery (NCD) has emerged as a vital area of research in machine learning and computer vision, aiming to identify novel classes in unlabeled datasets by leveraging knowledge from labeled datasets. Recent advances in NCD, especially after 2023, have focused on addressing key challenges such as imbalanced data, catastrophic forgetting, and improving the generalization capabilities of models. We examine how recent works integrate incremental learning, self-supervised techniques, and uncertainty quantification to enhance the discovery of novel classes. The role of generative models and transfer learning is also highlighted, particularly in domain-specific applications such as remote sensing, biomedical data, and synthetic aperture radar (SAR) imagery. Our review provides insights into the strengths, limitations, and future directions of NCD research, focusing on scalability, interpretability, and real-world applicability.