Linear Neighborhood Spread: A Way for Semi-Supervised Learning
Hui Xia He, Bo Chen, Jun Hai Guo · 2008
This paper is to introduce a novel semi-supervised learning algorithm named linear neighborhood spread (LNS), which is capable for learning manifold structures. Labeled and unlabeled data are represented as vertices in a weighted graph, and each data point is assumed can be linearly constructed from its neighborhood. Labels are spread through the edges, and the weighted graph is regarded as probabilistic transition matrix in the process of spread. In various experiments including synthetic data, digit and text classification, LNS showed promising performance.