Title :
Fast nonlinear dimension reduction
Author :
Kambhatla, Nandakishore ; Leen, Todd K.
Author_Institution :
Dept. of Comput. Sci. & Eng., Oregon Graduate Inst. of Sci. & Technol., Beaverton, OR, USA
Abstract :
A new algorithm for nonlinear dimension reduction is presented. The algorithm builds a piecewise linear model of the data. It provides compression that is superior to the globally linear model produced by principal component analysis. On several examples the piecewise linear model also provides compression that is superior to the global nonlinear model constructed by a five-layer, autoassociative neural network. The new algorithm trains significantly faster than the autoassociative network
Keywords :
data compression; neural nets; piecewise-linear techniques; compression; neural network; nonlinear dimension reduction; piecewise linear model; Computer science; Data analysis; Drives; Ear; Neural networks; Pattern recognition; Piecewise linear techniques; Principal component analysis; Redundancy; Vectors;
Conference_Titel :
Neural Networks, 1993., IEEE International Conference on
Conference_Location :
San Francisco, CA
Print_ISBN :
0-7803-0999-5
DOI :
10.1109/ICNN.1993.298730
