Abstract | ||
---|---|---|
The Universum data, deflned as a collection of "non- examples" that do not belong to any class of inter- est, have been shown to encode some prior knowledge by representing meaningful concepts in the same do- main as the problem at hand. In this paper, we ad- dress a novel semi-supervised classiflcation problem, called semi-supervised Universum, that can simultane- ously utilize the labeled data, unlabeled data and the Universum data to improve the classiflcation perfor- mance. We propose a graph based method to make use of the Universum data to help depict the prior infor- mation for possible classiflers. Like conventional graph based semi-supervised methods, the graph regulariza- tion is also utilized to favor the consistency between the labels. Furthermore, since the proposed method is a graph based one, it can be easily extended to the multi- class case. The empirical experiments on the USPS and MNIST datasets are presented to show that the pro- posed method can obtain superior performances over conventional supervised and semi-supervised methods. |
Year | Venue | Field |
---|---|---|
2008 | SDM | ENCODE,Graph,MNIST database,Pattern recognition,Computer science,Graph regularization,Artificial intelligence,Labeled data,Machine learning |
DocType | Citations | PageRank |
Conference | 24 | 1.36 |
References | Authors | |
9 | 4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Dan Zhang | 1 | 461 | 22.17 |
Jingdong Wang | 2 | 4198 | 156.76 |
Fei Wang | 3 | 2139 | 135.03 |
Changshui Zhang | 4 | 5506 | 323.40 |