计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (24): 128-135.DOI: 10.3778/j.issn.1002-8331.1808-0341

• 模式识别与人工智能 • 上一篇    下一篇

基于核极限学习机的标记分布学习

王一宾,田文泉,程玉胜,裴根生   

  1. 安庆师范大学 计算机与信息学院,安徽 安庆 246011
  • 出版日期:2018-12-15 发布日期:2018-12-14

Label distribution learning based on kernel extreme learning machine

WANG Yibin, TIAN Wenquan, CHENG Yusheng, PEI Gensheng   

  1. School of Computer and Information, Anqing Normal University, Anqing, Anhui 246011, China
  • Online:2018-12-15 Published:2018-12-14

摘要: 标记分布学习作为一种新的学习范式,利用最大熵模型构造的专用化算法能够很好地解决某些标记多样性问题,但是计算量巨大。基于此,引入运行速度快、稳定性更高的核极限学习机模型,提出基于核极限学习机的标记分布学习算法(KELM-LDL)。首先在极限学习机算法中通过RBF核函数将特征映射到高维空间,然后对原标记空间建立KELM回归模型求得输出权值,最后通过模型计算预测未知样本的标记分布。与现有算法在各领域不同规模数据集的实验表明,实验结果均优于多个对比算法,统计假设检验进一步说明KELM-LDL算法的有效性和稳定性。

关键词: 标记分布学习, 极限学习机, 回归拟合, 核函数

Abstract: Label distribution learning, as a new learning paradigm, uses the specialized algorithm constructed by the maximum entropy model to solve many problems with label ambiguity well, but it is extremely computational intensive. Based on this, the kernel extreme learning machine model with fast running speed and high stability is introduced, and a label distribution learning algorithm based on this model is proposed. Firstly, the features are mapped to high-dimensional space by RBF kernel function in extreme learning machine algorithm. Then, the KELM regression model is established to obtain the output weight for the original label space. Finally, this model is used to predict the label distribution of unknown samples. The comparison between KELM-LDL and the existing algorithm in different sizes of data in various fields show that experimental results of this algorithm are superior to other comparison algorithms, and statistical hypothesis testing further illustrates the effectiveness and stability of the KELM-LDL algorithm.

Key words: label distribution learning, extreme learning machine, regression fitting, kernel function