Softmax Regression by Using Unsupervised Ensemble Learning

Logistic regression and softmax regression are effective ways to handle two-class and multi-class classification problems, but they have to need groudtruth labels. Labeled instances however are often difficult, expensive, and time consuming to obtain. It poses huge challenges to classification. Unsupervised learning and ensemble learning often are good ways to get labels. In this study, the objective function is constructed based on softmax regression and unsupervised learning. For the inference, we propose evaluation index C to measure the confidence of unsupervised ensemble learning. Furthermore, a corresponding SRuel algorithm is designed. Finally, extensive experiments demonstrate the high performance of the proposed algorithm.