Closed Set

Closed-set classification, a core problem in machine learning, assumes all data belongs to known classes during both training and testing. Current research focuses on extending this to open-set scenarios, where unseen classes may appear at test time, necessitating robust methods for both classification and rejection of unknown data. This involves developing novel loss functions, adapting existing architectures like convolutional prototype networks, and employing techniques such as self-learning and background-class regularization to improve performance. The ability to reliably handle open-set problems is crucial for numerous applications, including biometrics, forensic science, and medical diagnosis, where encountering unknown data is common.