Graph Constuction Based on Fast Low-Rank Representation in Graph-Based Semi-Supervised Learning

摘要

According to the present invention, a method for constructing a graph in graph-based semi-supervised learning comprises the steps: (a) receiving a data set (X) and a nearest neighbor number (k); (b) generating a k-NN graph using the data set (X), and then calculating L; (c) decomposing X using SkinnySVD; (d) obtaining βL using SVD; (e) updating J, W, and Q while satisfying a predetermined condition; and (f) obtaining an optimal solution by calculating Y_1 and Y_2, such that it is possible to construct a graph based on a fast low-rank representation algorithm. According to the present invention, the method for constructing a graph is based on a fast low-rank representation and configured to introduce additional constraints to an underlying optimization goal and optimize the underlying optimization goal such that it is possible to quickly acquire a better solution.