嵌入拒识代价的投票式极限学习机（英文）

Voting extreme learning machine with rejection

极限学习机是近几年发展起来的一种单隐层前馈神经网络.通过训练多个独立的ELM,V-ELM不仅提高了ELM的分类精度,同时很好地解决了ELM不稳定的特性.在V-ELM中,需要计算一个样本属于每一类的概率,将样本分类为概率最大的那一类.然而,当遇到最大的两个概率相等或者相差不大的情况下,都对应着非常大的误分类概率,为了解决这一问题,在论文中引入了对样本的拒识决策,并将该方法命名为嵌入拒识的投票式极限学习机.

Extreme learning machine （ELM） was developed in recent years for a single hidden layer feed forward neural network. Through the training of multiple independent ELM rather than a single ELM, voting extreme learning machine （V-ELM） not only improved the classification performance by reducing the number of misclassification samples, but also reduced the inconsistency of ELM in the learning process. In V-ELM algorithm, it need to calculate the probabilities one sample belongs to the expected class labels, and then classified the sample to the class which has the largest probability the sample belonging to. However, when it comes to the phenomenon that the two Or more probabilities one sample belong to were equal or very close in a certain extent then the classification risk was great. In order to solve this problem, we put forward in this paper a new algorithm called embed rejection to voting extreme learning machine （RV-ELM）.