Semi-Supervised Learning Based on Extreme Learning

Semi-supervised learning with growing self-organizing map (GSOM) is used in many applications, such as clustering. The main challenges in the Semi-supervised GSOM are calculating parameters such as shape and structure of clustering layer, activation level, and weights of classifier layer. Current approaches use initiative methods with a local look have trying to determine these parameters; which its effect, the results of these algorithms is highly dependent on the conditions. This paper studies a semi-supervised learning method based on GSOM and extreme learning for the first time. The proposed method, without the direct calculation of the GSOM parameters and using the extreme learning determines label of each data. Error resulted from the feedback system is used to optimize extreme learning and GSOM. In this paper, in addition to investigating the convergence analysis of the proposed method, sequential extreme learning is also provided for semi-supervised GSOM. Experiments conducted on online and partially labeled data show that the proposed method has a relative advantage in terms of accuracy on semi-supervised GSOM.