Semi-supervised Learning by Maximizing Smoothness
Dengyong Zhou, Olivier Bousquet, Thomas Navin Lal, Jason Weston, Max Planck · 2004
We consider the general problem of learning from labeled and unlabeled data. Given a set of points, some of them are labeled, and the remaining points are unlabeled. The goal is to predict the labels of the unlabeled points. Any supervised learning algorithm can be applied to this problem, for instances, Support Vector Machines (SVMs). The problem of our interest is if we can implement a classier which uses the unlabeled data information in some way and has higher accuracy than the classiers which use the labeled data only. Here we propose a simple algorithm to utilize the unlabeled data. The basic idea is to construct the classifying function which sucien tly smooth with respect to the intrinsic global structure collectively revealed by known labeled and unlabeled points. The method yields encouraging experimental results on a number of classication problems and demonstrates eectiv e use of unlabeled data.