作　　者： (赵德海）; (江兵兵）;

机构地区： 中国科学技术大学,计算机科学与技术学院,安徽省合肥市230026

出　　处： 《电子技术（上海）》 2017年第8期19-24,共6页

摘　　要： 由黄广斌等人提出的极速学习机(ELM)模型成为近年来机器学习领域的热门研究方向,因其具有较低的计算复杂度,被广泛应用于学习多层神经网络中的隐层参数。但是,当前基于极速学习机框架的分类算法在稀疏性和分类准确率方面表现欠佳。本文在稀疏贝叶斯和极速学习机框架的基础上,提出一种基于稀疏贝叶斯的半监督极速学习机分类算法。该算法通过在输出层的权值参数上定义稀疏流形先验,充分利用了样本数据的局部信息,提高了模型的分类准确率。同时,通过在学习阶段自动修剪冗余隐层节点,降低了极速学习机的精度对隐层节点数量的敏感性。多个数据集上的实验结果表明:本文算法在分类准确率方面可以和当前主流的半监督分类器进行比较,同时模型具有良好的稀疏性。 Extreme learning machine（ELM） has become a hot topic in the last few years. Due to its lower computational complexity, ELM has been widely used to learn hidden layer parameters of multilayer neural networks. However, ELM-based classification algorithms cannot perform well in terms of sparseness and classification accuracy. In this paper, a sparse Bayesian based semi-supervised learning algorithm is proposed, which is based on the sparse Bayesian inference and ELM. This new algorithm is called sparse Bayesian semi-supervised ELM（SBSSELM）. The proposed algorithm can make full use of the local information and simultaneously improve the classification accuracy through introducing a manifold prior on weights of the output layer. In addition, by automatically pruning the redundant hidden neurons in the learning procedure, it can obtain a relatively insensitive model to the number of hidden neurons. The experiments on various data sets demonstrate that SBSSELM can give a sparse model with comparable classification accuracy in comparison to some state-of-the-art semi-supervised methods.

关 键 词： 半监督学习 稀疏贝叶斯学习 分类 极速学习机 流形正则化