Abstract: Extreme learning machine is widely used in the field of face recognition. The traditional extreme learning algorithm easily leads to the problem of insufficient learning process due to training on a small number of labeled samples. At the same time, the intrinsic geometry of the sample is often ignored in the learning process, which affects its classification ability for face recognition. Inspired by the idea of manifold learning, a neighborhood preserving extreme learning algorithm is proposed. The algorithm preserves the most essential structure of data and discriminant information of intra-class data, and minimizes the intra-class divergence matrix to improve the classification performance of the extreme learning machine. The experimental  results on face datasets show that the algorithm has a higher accuracy of face recognition than other algorithms and is more effective in classification and recognition.

Key words: extreme learning machine, manifold learning, neighborhood preserving embedding, geometric structure

摘要： 极限学习机广泛应用于人脸识别领域。传统的极限学习机算法因在少量标签样本上进行训练，容易发生学习过程不充分问题，同时在学习过程中往往忽略了样本内在的几何结构，影响其对人脸识别的分类能力。受流形学习思想的启发，提出一种邻域保持极限学习机算法。该算法保持数据最本质的结构和同类数据的判别信息，利用最小化类内散度矩阵来提高极限学习机整体的分类性能。通过人脸数据集上的多次实验结果表明，该算法的人脸识别准确率高于其他算法，更能有效地进行分类识别。

关键词: 极限学习机, 流形学习, 近邻保持嵌入, 几何结构