基于多输出极限学习机的快速一致性分类器

一致性分类器是建立在一致性预测基础上的分类器,其输出结果具有很高的可靠性,但由于计算框架的限制,学习的时间往往较长.为了加快学习速度,首次将一致性预测与多输出极限学习机相结合,提出基于两者的快速一致性分类算法.该算法利用了极限学习机,能够快速计算样本标签的留一交叉估计的特性,极大地加快了学习速度.算法复杂度分析表明,所提算法的计算复杂度与多输出极限学习机的算法复杂度相同,该算法继承了一致性预测的可靠性特征,即预测的错误率能够被显著性水平参数所控制.在10个公共数据集上的对比实验表明,所提算法具有极快的计算速度,且与其他常用一致性分类器相比,该算法的平均预测标签个数在某些数据集上更少,预测结果更有效.

A fast conformal classifier based on multi-output extreme learning machine

A conformal classifier is a conformal prediction based classifier. Although the prediction is highly valid, the learning time of conformal classifiers is often very long due to the limitation of the computational framework. To make the conformal classifier learn faster, this paper firstly proposes an algorithm combining the conformal prediction with the multi-output extreme learning machine whose leave-one-out predictions on the training set can be computed efficiently. From the analysis of algorithm complexity, the computational complexity of the proposed algorithm is equivalent to that of the multi-output extreme learning machine. The experiments on ten public data sets show that the proposed our algorithm has fast computation speed and inherits the property of validity from conformal prediction, whose prediction error can be controlled by the significance level. The average number of labels per prediction of the proposed algorithm is lower than that of other common conformal classifiers on some data sets, which shows that of is more efficient in some applications.