Performance analysis of adaptive eigenanalysis algorithms

Authors:
V. Solo;Xuan Kong
Affiliations:
Dept. of Stat., Macquarie Univ., North Ryde, NSW;-
Venue:
IEEE Transactions on Signal Processing
Year:
1998

Citing 0
Cited 4

Adaptive algorithms for solving generalized eigenvalue signal enhancement problems

Signal Processing
Recursive total instrumental-variable algorithm for solving over-determined normal equations and its applications

Signal Processing
Convergence analysis of a class of Hyvärinen-Oja's ICA learning algorithms with constant learning rates

IEEE Transactions on Signal Processing
Adaptive multiple minor directions extraction in parallel using a PCA neural network

Theoretical Computer Science

Quantified Score

Hi-index	35.69

Visualization

Abstract

We present a rigorous analysis of several popular forms of short memory adaptive eigenanalysis algorithms using a stochastic averaging method. A first-order analysis shows that the algorithms do not have any equilibrium points despite published claims to the contrary. Through averaging analysis, we show that they hover around an appropriate eigenvector. A second-order analysis is also given without the Gaussian noise assumption, and our results greatly outperform an earlier approximation in the literature. The second-order analysis has been of much interest in the offline study but, in the dynamic adaptive case, is uncommon