点击次数：

论文类型：期刊论文

发表时间：2015-09-01

发表刊物：Journal of Information and Computational Science

收录刊物：EI、Scopus

卷号：12

期号：13

页面范围：4951-4958

ISSN号：15487741

摘要：Original ELM-AE algorithm combines Extreme Learning Machine (ELM) with Auto Encoder (AE) to map data to a nonlinear high-dimensional space. This procedure can extract better sample characteristics when solving unsupervised clustering problems. ELM-AE adopts a solution procedure which is similar to ELM to improve the solution speed. However, ELM-AE has some problems. The solution of the system is not stable because of the random input-layer weights. Meanwhile, if the number of hidden layers is small, model fitting effect will be significantly reduced. To solve the above problems, we propose a Robust ELM-AE (RELM-AE) clustering algorithm. This method trains Restricted Boltzmann Machine (RBM) to get stable input-layer weights, and then minimizes the output error to obtain the output-layer weights. Compared with the original model of ELM-AE, this method achieves more stable and accurate results, which makes it more suitable for solving unsupervised clustering problems. ?, 2015, Journal of Information and Computational Science. All right reserved.