Relaxational metric adaptation and its application to semi-supervised clustering and content-based image retrieval

Chang, Hong; Yeung, Dit–Yan; Cheung, William K.

doi:10.1016/j.patcog.2006.04.006

Cited by 21 publications

(12 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Some experiments on the synthetic, UCI, and MNIST datasets demonstrated the superiority of our method over some existing linear and non-linear metric learning methods. As the future work, we will apply the proposed methods on other real-world problems such as content-based image retrieval [5].…”

Section: Resultsmentioning

confidence: 99%

Probabilistic non-linear distance metric learning for constrained clustering

Babagholami-Mohamadabadi

Zarghami

Pourhaghighi

et al. 2013

Proceedings of the 4th MultiClust Workshop on Multiple Clusterings, Multi-View Data, and Multi-Source Knowledge-Driven Clusteri

View full text Add to dashboard Cite

Distance metric learning is a powerful approach to deal with the clustering problem with side information. For semisupervised clustering, usually a set of pairwise similarity and dissimilarity constraints is provided as supervisory information. Although some of the existing methods can use both equivalence (similarity) and inequivalence (dissimilarity) constraints, they are usually limited to learning a global Mahalanobis metric (i.e., finding a linear transformation). Moreover, they find metrics only according to the data points appearing in constraints, and cannot utilize information of other data points. In this paper, we propose a probabilistic metric learning algorithm which uses information of unconstrained data points (data points which do not appear in neither positive nor negative constraints) along with both positive and negative constraints. We also kernelize our metric learning method based on the kernel trick which provides a non-linear version of the learned metric. Experimental results on synthetic and real-world data sets demonstrate the effectiveness of the proposed metric learning algorithm.

show abstract

Section: Resultsmentioning

confidence: 99%

Probabilistic non-linear distance metric learning for constrained clustering

Babagholami-Mohamadabadi

Zarghami

Pourhaghighi

et al. 2013

Proceedings of the 4th MultiClust Workshop on Multiple Clusterings, Multi-View Data, and Multi-Source Knowledge-Driven Clusteri

View full text Add to dashboard Cite

show abstract

“…As another direction, we will incorporate dissimilarity constraints into the methods to further improve the metric learning performance. Moreover, we will explore the application of the proposed methods to other real-world problems such as content-based image retrieval [7], [8], [9].…”

Section: Discussionmentioning

confidence: 99%

A Kernel Approach for Semisupervised Metric Learning

Yeung

Chang

2007

IEEE Trans. Neural Netw.

View full text Add to dashboard Cite

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 semi-supervised metric learning based on pairwise similarity or dissimilarity information. In this paper, we propose a kernel approach for semi-supervised 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 closed-form 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 real-world data show that this new kernel approach is promising for nonlinear metric learning.

show abstract

“…Examples on such situations where a considerable better classification accuracy have been obtained by applying some other distance measure have been reported in several articles i.e. [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [31]. However, typically in these types of studies one has simply tested with a few different distance measure for classifying the data set at hand.…”

Section: Introductionmentioning

confidence: 99%

Differential evolution classifier with optimized distance measures from a pool of distances

Koloseni

Lampinen

Luukka

2012

2012 IEEE Congress on Evolutionary Computation

View full text Add to dashboard Cite

In this article we propose a differential evolution based nearest prototype classifier with extension to selecting the applied distance measure from a pool of alternative measures optimally for the particular data set at hand. The proposed method extends the earlier differential evolution based nearest prototype classifier by extending the optimization process to cover also the selection of distance measure instead of optimizing only the parameters related with a preselected and fixed distance measure. Now the optimization process is seeking also for the best distance measure providing the highest classification accuracy over the selected data set. It has been clear for some time that in classification, the usual euclidean distance measure is sometimes not the best possible choice. Still usually not much has been done for it, and in many cases where some consideration to this problem is given, there has only been testing with a couple of alternative distance measures to find which one provides the highest classification accuracy over the current data set. In this paper we attempt to take one step further by not only enumerating a couple of alternative distance measures, but applying a systematic optimization process to select the best distance measure from a pool of multiple alternative distance measures. In parallel, within the same optimization process, the optimal parameter values related to each alternative distance measures are determined as well as the optimal class prototype vectors for the given data. The empirical results represented are indicating that with several data sets the optimal distance measure is some other measure than the most commonly applied euclidean distance. The results are also suggesting that from the classification accuracy point of view the proposed global optimization approach has high potential in solving classification problems of the studied type. Perhaps the most generally applicable conclusion from our results is, that emphasizing of selection of distance measure is more important to classification accuracy that it has been commonly believed so far.

show abstract

Relaxational metric adaptation and its application to semi-supervised clustering and content-based image retrieval

Cited by 21 publications

References 25 publications

Probabilistic non-linear distance metric learning for constrained clustering

Probabilistic non-linear distance metric learning for constrained clustering

A Kernel Approach for Semisupervised Metric Learning

Differential evolution classifier with optimized distance measures from a pool of distances

Contact Info

Product

Resources

About