Abstract
Some configurations to be discussed • Laplacian problems: (1) determining the size of the analysis; (2) determining the number of neighboring points; (3) multi-scale processing transactions; (4) for noise and outliers .
• In this paper to calculate the similarity, this non-parametric method can reduce the computational complexity while improving robustness by determining the constraints of sparse representation.
1 Background and Motivation
• generate a lot of algorithms based on graph theory and applications: (1) clustering algorithm; (2) dimensionality reduction algorithm; (3) semi-supervised learning algorithm; (4) the ranking algorithm.