Results 1 
4 of
4
SemiSupervised Metric Learning Using Pairwise Constraints
"... Distance metric has an important role in many machine learning algorithms. Recently, metric learning for semisupervised algorithms has received much attention. For semisupervised clustering, usually a set of pairwise similarity and dissimilarity constraints is provided as supervisory information. ..."
Abstract

Cited by 13 (0 self)
 Add to MetaCart
Distance metric has an important role in many machine learning algorithms. Recently, metric learning for semisupervised algorithms has received much attention. For semisupervised clustering, usually a set of pairwise similarity and dissimilarity constraints is provided as supervisory information. Until now, various metric learning methods utilizing pairwise constraints have been proposed. The existing methods that can consider both positive (mustlink) and negative (cannotlink) constraints find linear transformations or equivalently global Mahalanobis metrics. Additionally, they find metrics only according to the data points appearing in constraints (without considering other data points). In this paper, we consider the topological structure of data along with both positive and negative constraints. We propose a kernelbased metric learning method that provides a nonlinear transformation. Experimental results on synthetic and realworld data sets show the effectiveness of our metric learning method. 1
Learning Assignment Order of Instances for Constrained Kmeans Clustering Algorithm
, 2007
"... Constrained Kmeans clustering algorithm with instancelevel background knowledge can often achieve a better clustering solution when compared with the one obtained by traditional unsupervised Kmeans clustering algorithm. However, constrained Kmeans clustering algorithm suffers from the problem of ..."
Abstract

Cited by 2 (0 self)
 Add to MetaCart
Constrained Kmeans clustering algorithm with instancelevel background knowledge can often achieve a better clustering solution when compared with the one obtained by traditional unsupervised Kmeans clustering algorithm. However, constrained Kmeans clustering algorithm suffers from the problem of the sensitivity to the assignment order of instances. To the best knowledge of the authors’, very few work has been done on learning a good assignment order of instances for constrained Kmeans clustering algorithm. This paper explores the above problem and proposes a novel assignment order learning algorithm for constrained Kmeans clustering algorithm, termed as Clustering Uncertainty based assignment order Learning Algorithm (UALA). UALA ranks all instances in the data set according to their clustering uncertainty that is calculated by using the ensembles of multiple clustering algorithms. We test UALA on several real data sets with artificial instancelevel constraints. Experimental results demonstrate that UALA is able to identify a good assignment order of instances for constrained Kmeans clustering algorithm, therefore can significantly improve the accuracy of constrained Kmeans clustering algorithm. Index Terms Clustering analysis, constrained Kmeans clustering algorithm, semisupervised clustering with instancelevel constraints I.
A Kernel Approach for SemiSupervised Metric Learning
"... Abstract — While distance function learning for supervised learning tasks has a long history, extending it to learning tasks with weaker supervisory information has only been studied recently. In particular, some methods have been proposed for semisupervised metric learning based on pairwise simila ..."
Abstract
 Add to MetaCart
(Show Context)
Abstract — While distance function learning for supervised learning tasks has a long history, extending it to learning tasks with weaker supervisory information has only been studied recently. In particular, some methods have been proposed for semisupervised metric learning based on pairwise similarity or dissimilarity information. In this paper, we propose a kernel approach for semisupervised metric learning and present in detail two special cases of this kernel approach. The metric learning problem is thus formulated as an optimization problem for kernel learning. An attractive property of the optimization problem is that it is convex and hence has no local optima. While a closedform solution exists for the first special case, the second case is solved using an iterative majorization procedure to estimate the optimal solution asymptotically. Experimental results based on both synthetic and realworld data show that this new kernel approach is promising for nonlinear metric learning. Index Terms — metric learning, kernel learning, semisupervised learning, clustering. I.
Two Phase Semisupervised Clustering Using Background Knowledge
"... Abstract. Using background knowledge in clustering, called semiclustering, is one of the actively researched areas in data mining. In this paper, we illustrate how to use background knowledge related to a domain more efficiently. For a given data, the number of classes is investigated by using the ..."
Abstract
 Add to MetaCart
(Show Context)
Abstract. Using background knowledge in clustering, called semiclustering, is one of the actively researched areas in data mining. In this paper, we illustrate how to use background knowledge related to a domain more efficiently. For a given data, the number of classes is investigated by using the mustlink constraints before clustering and these mustlink data are assigned to the corresponding classes. When the clustering algorithm is applied, we make use of the cannotlink constraints for assignment. The proposed clustering approach improves the result of COP kmeans by about 10%. 1