Co-MLM: a SSL algorithm based on the Minimal Learning Machine

Abstract

Semi-supervised learning is a challenging topic in machine learning that has attracted much attention in recent years. The availability of huge volumes of data and the work necessary to label all these data are two of the reasons that can explain this interest. Among the various methods for semisupervised learning, the co-training framework has become popular due to its simple formulation and promising results. In this work, we propose Co-MLM, a semi-supervised learning algorithm based on a recently supervised method named Minimal Learning Machine (MLM), built upon co-training framework. Experiments on UCI data sets showed that Co-MLM has promising performance in compared to other co-training style algorithms.