Theoretical Analysis of a Rigid Coreset Minimum Enclosing Ball Algorithm for Kernel Regression Estimation

Authors:
Xunkai Wei;Yinghong Li
Affiliations:
Beijing Aeronautical Technology Research Center, Beijing, China 100076 and Air Force Engineering University, Xian, 710038;Air Force Engineering University, Xian, 710038
Venue:
ISNN '08 Proceedings of the 5th international symposium on Neural Networks: Advances in Neural Networks
Year:
2008

Citing 8
Cited 0

Support vector domain description

Pattern Recognition Letters - Special issue on pattern recognition in practice VI
Approximate clustering via core-sets

STOC '02 Proceedings of the thiry-fourth annual ACM symposium on Theory of computing
Smaller core-sets for balls

SODA '03 Proceedings of the fourteenth annual ACM-SIAM symposium on Discrete algorithms
Approximate minimum enclosing balls in high dimensions using core-sets

Journal of Experimental Algorithmics (JEA)
A tutorial on support vector regression

Statistics and Computing
Core Vector Machines: Fast SVM Training on Very Large Data Sets

The Journal of Machine Learning Research
Enclosing machine learning: concepts and algorithms

Neural Computing and Applications
Generalized Core Vector Machines

IEEE Transactions on Neural Networks

Quantified Score

Hi-index	0.00

Visualization

Abstract

A rigid coreset minimum enclosing ball training machine for kernel regression estimation was proposed. First, it transfers the kernel regression estimation machine problem into a center-constrained minimum enclosing ball representation form, and subsequently trains the kernel methods using the proposed MEB algorithm. The primal variables of the kernel methods are recovered via KKT conditions. Then, detailed theoretical analysis and main theoretical results of our new algorithm are given. It can be concluded that our proposed MEB training algorithm is independent of sample dimension and the time complexity is linear in sample numbers, which greatly cuts down the complexity level and is expected to speedup the learning process obviously. Finally, comments about the future development directions are discussed.