Data Models and Structures of Kernels of DR
In the dimensionality reduction processing, observed data have two different types. In the first type, the data set consists of high-dimensional vectors, which represent the objects of interest. In the second type, the data describe the similarities (or dissimilarities) of objects that cannot be digitized or hidden. The output of a DR processing with an input of the first type is a low-dimensional data set, having the same cardinality as the input and preserving the similarities of the input. When the input is of the second type, the output is a configuration of the input similarities. In Section 1 of this chapter, we discuss the models of input data and the constraints on output data. In Sections 2, we discuss the construction of the kernels in DR methods.
KeywordsDimensionality Reduction Local Linear Embedding Euclidean Distance Matrix Semisupervised Learning Principal Component Analysis Project
Unable to display preview. Download preview PDF.
- Carreira-Perpiñán, M.Á.: A review of dimension reduction techniques. Tech. Rep. CS-96-09, Dept. of Computer Science, University of Sheffield, UK (1996).Google Scholar
- Lu, F., Keleş, S., Lin, Y., Wright, S., Wahba, G.: Kernel regularization and dimension reduct. Tech. rep., Department of Statistics, University of Wisconsin (2006).Google Scholar
- Saul, L.K., Weinberger, K.Q., Sha, F., Ham, J., Lee, D.D.: Spectral methods for dimensionality reduction. In: B. Schóelkopf, O. Chapelle, A. Zien (eds.) Semisupervised learning. MIT Press (2005).Google Scholar
- Xiao, L., Sun, J., Boyd, S.P.: A duality view of spectral methods for dimensionality reduction. In: W.W. Cohen, A. Moore (eds.) Machine Learning: Proceedings of the Twenty-Third International Conference, ACM International Conference Proceeding Series, vol. 148, pp. 1041–1048. ICML, Pittsburgh, Pennsylvania, USA (2006).Google Scholar