Discovery of Unknown Classes in Constrained Semi-Supervised Learning

Supervised classification models have proved their significance in research and are utilized in many key areas. However, there are some limitations for these methods. In supervised learning, adequate labelled training data are required for each class. The learned classifier can classify only the data belonging to classes for which training data is available and lacks the ability to handle the data of unseen classes. However, in real time scenarios, there may not be adequate training data for all the classes. There could also be scenarios where the classes not available in the training data appearing in the testing data. The learner is expected to have the capability to uncover the class labels for the data belonging to unseen classes. To handle this situation, methods under different learning paradigms are proposed. In this work, various semi supervised learning techniques that can handle hidden classes(clusters) are evolved. Variant of Semi Supervised EM Based-K means and Naïve Bayes are discussed.