Self labeling online sequential extreme learning machine and it’s application

Abstract

In the field of big data machine learning, the data volume is large, but the labeled data is few. Due to this, it may lead to that the distribution of labeled data (source domain) is not similar to that of unlabeled data (target domain). In traditional machine learning field, this problem is a kind of transfer learning problems. To address this problem, a self labeling online sequential extreme learning machine is presented, which is called SLOSELM. Firstly, an ELM classifier is trained on the labeled training dataset of the source domain. Secondly, the unlabelled dataset of the target domain is classified by the ELM classifier. In the third step, the high confident samples are selected and the OSELM is employed to update the original ELM classifier. Tested on the real-world image dataset and the daily activity dataset, the results show that our algorithm performs well.