Abstract: Nonlinear manifold learning methods for dimensionality reduction are widely applied to face recognition, intrusion detection and sensor networks, etc. However, few manifold learning algorithms can deal effectively with sparse data. This paper proposes a sparse embedding algorithm based on conceptual framework of Local Linear Embedding algorithm（LLE）, which can achieve the purpose of extensive overlapping of information by expanding and strengthening local neighborhood information in the case of sparse samples. The experimental results on sparse artificial and face datasets show that the proposed algorithm generates a better embedding and classification result.

Key words: nonlinear manifold learning, dimensionality reduction, Local Linear Embedding（LLE）

摘要： 非线性流形学习降维方法已经被广泛应用到人脸识别、入侵检测以及传感器网络等领域。然而，能够有效处理稀疏数据的流形学习算法很少。基于局部线性嵌入（LLE）算法的思想框架，提出一种扩大局部邻域的稀疏嵌入算法，通过对局部区域信息加强，使得在样本较少的情况下，达到丰富重叠信息的目的。在稀疏的人工和人脸数据集上的实验结果表明，所提算法产生了较好的嵌入及分类结果。

关键词: 非线性流形学习, 降维, 局部线性嵌入