基于流形学习的极端学习机及其在人脸识别中的应用

Using manifold learning extreme learning machine for face recognition

摘要:

极端学习机以其快速高效和良好的泛化能力在模式识别领域得到了广泛应用,然而现有的ELM及其改进算法并没有充分考虑到数据维数对ELM分类性能和泛化能力的影响,当数据维数过高时包含的冗余属性及噪音点势必降低ELM的泛化能力,针对这一问题本文提出一种基于流形学习的极端学习机,该算法结合维数约减技术有效消除数据冗余属性及噪声对ELM分类性能的影响,为验证所提方法的有效性,实验使用普遍应用的图像数据,实验结果表明本文所提算法能够显著提高ELM的泛化性能。

Abstract:

Extreme learning machine has been widely used for its fast, efficient and good generalization ability in pattern recognition. But existing ELM and its improved algorithms do not adequately consider the impact of data dimension for classification performance and generalization ability of ELM, the redundant attributes and noise will reduce the generalization ability of ELM which contains when data dimension is too high. Aiming at this problem, this paper proposed a kind of extreme learning machine based on manifold learning. The algorithm combines dimensionality reduction technology, effectively eliminate the impact of data redundancy attribute and noise for classification performance of ELM. In order to verify the effectiveness of the proposed method, experimental use of widely used image data, experimental results show that the proposed algorithm can significantly improve the generalization performance of ELM.