Weighted Kernel Isomap for Data Visualization and Pattern Classification

Authors:
Rui-Jun Gu;Wen-Bo Xu
Affiliations:
School of Information Technology, Southern Yangtze University, Wuxi 214122, China;School of Information Technology, Southern Yangtze University, Wuxi 214122, China
Venue:
Computational Intelligence and Security
Year:
2007

Citing 4
Cited 0

Efficient Pattern Recognition Using a New Transformation Distance

Advances in Neural Information Processing Systems 5, [NIPS Conference]
Laplacian Eigenmaps for dimensionality reduction and data representation

Neural Computation
Robust kernel Isomap

Pattern Recognition
Supervised nonlinear dimensionality reduction for visualization and classification

IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics

Quantified Score

Hi-index	0.00

Visualization

Abstract

Dimensionality reduction is an important task in pattern recognition and data mining. Isomap is a representative of manifold learning approaches for nonlinear dimensionality reduction. However, Isomap is an unsupervised learning algorithm and has no out-of-sample ability. Kernel Isomap (KIsomap) is an improved Isomap and has a generalization property by utilizing kernel trick. At first, considering class label, a Weighted Euclidean Distance (WED) is designed. Then, WED based kernel Isomap (WKIsomap) is proposed. As a supervised learning algorithm, WKIsomap can not only be used in data visualization, but also applied to feature extraction for pattern recognition. The experimental results show that WKIsomap is more robust than Isomap and KIsomap in data visualization. Moreover, when noise is added into data, WKIsomap based classifiers are more robust to noise than KIsomap based ones.