Semi-Supervised Classification Using Linear Neighborhood Propagation

We consider the general problem of learning from both labeled and unlabeled data. Given a set of data points, only a few of them are labeled, and the remaining points are unlabeled. Our goal is to predict the labels of both the unlabeled and new out-of-sample data points. Based on the assumption that the labels of each data can be linearly reconstructed from its neighbors' labels, we develop a novel algorithm to achieve this goal. The new method, called Linear Neighborhood Propagation (LNP), aims at learning the whole neighborhood structure of each data object, and use these structures to "propagate" the labels of the labeled points to the remaining unlabeled points. The main procedure of our approach is composed of two steps, first it computes the local neighborhood geometry of the dataset by solving a constrained least squares problem, then the computed geometry will be used to predict the labels of unlabeled data by solving a linearly constrained quadratic optimizati...

• Type: Conference
• Year: 2006
• Where: CVPR
• Authors: Fei Wang, Changshui Zhang, Helen C. Shen, Jingdong Wang
• Keywords: Computer Vision, CVPR 2006, Linear Neighborhood Propagation, Local Neighborhood Geometry, Out-of-sample Data Points, Unlabeled Data, Unlabeled Points