Robust Semi-Supervised Growing Self-Organizing Map

Semi-Supervised Growing Self Organizing Map (SSGSOM) is one of the best methods for online classification with partial labeled data. Many parameters can affect the performance of this method. The structure of GSOM network, activation degree and learning approach are the most important factors in SSGSOM. In this paper, a comprehensive robust mathematical formulation of the problem is proposed and then half quadratic (HQ) is used to solve it. Furthermore, an adaptive method is proposed to adjust activation degree optimally to improve the performance of SSGSOM. The results are reported on a variety of synthetic and UCI datasets and in the noisy conditions, which show superiority and robustness of the proposed method compared with the state of the art approaches.