点击次数：

论文类型：会议论文

发表时间：2009-07-13

收录刊物：EI、Scopus

卷号：1

页面范围：357-361

摘要：Kernel density estimation (KDE) is a powerful and popular method of estimating the probability density function of a random sample. However, it needs high-computational requirement when large data samples are available. In this paper, a two stage density estimation method based on extreme learning machine (ELM) is proposed, which aims to estimate the density fonction with acceptable computational cost and accuracy. At the first stage, an ELM with empirical distribution function as the desired response is developed, and the output weight vector is obtained via pseudo inverse method. Then, at the second stage the hidden activation functions of ELM are replaced by some new designed ones, with a linear mapping between the hidden activation functions used in the first stage and the new ones. Finally, data sample can be imported into the model, and the output will be the probability density to be estimated. Experiments are done on three synthetic data, and the results compared with the KDE and some other state-of-the-art sparse kernel density (SKD) estimation techniques demonstrate the effectiveness of the proposed method.