A semi-supervised learning approach for soft labeled data
In some machine learning applications using soft labels is more useful and informative than crisp labels. Soft labels indicate the degree of membership of the training data to the given classes. Often only a small number of labeled data is available while unlabeled data is abundant. Therefore, it is important to make use of unlabeled data. In this paper we propose an approach for Fuzzy-Input Fuzzy-Output classification in which the classifier can learn with soft-labeled data and can also produce degree of belongingness to classes as an output for each pattern. Particularly, we investigate the case where only a few soft labels are available and data can be represented by different views. We investigate two semi-supervised multiple classifier frameworks for this classification purpose. Results show that semi supervised multiple classifiers can improve the performance of fuzzy classification by making use of the unlabeled data. © 2010 IEEE.