Locally linear embedding preserving local neighborhood
Tingquan Deng, Jinyan Liu, Ning Wang · 2016
Dimensionality reduction is an important issue in information processing and has popular applications in many fields, where locally linear embedding (LLE) is widely used due to accuracy and simple to implement. However, LLE is lack of robustness, and sensitive to local structure that can't preserve neighborhood character sometimes. Instead, Laplacian eigenmaps (LE) can overcome these weaknesses. In this paper, a new dimensionality reduction algorithm, called locally linear embedding preserving neighborhood (NLLE) is proposed. It takes the advantage of LLE and LE, and can keep the intrinsic character of high-dimensional data. Several experiments are employed to confirm the effectiveness and robustness of the algorithm.