Adaptive algorithms for first principal eigenvector computation
The paper is really devoted to applications of simple iterative algorithms for finding approximations \(w_k\) of the first principal eigenvector of an unknown matrix \(A\) when only its approximations \(A_k \) are given. These \(A_k\) are generated by a sequence of random observation vectors \(x_k\in {\mathbb R}^n\); for example, \(A_k\equiv x_kx_k^T\), \(A_k\equiv \beta A_{k-1}+k^{-1}(x_kx_k^T-\beta A_{k-1}).\) Several gradient methods of the type \(w_{k+1}=w_k+\tau_kr(w_k,A_k)\) are considered with different objective functions; some of them deal with the Rayleigh quotient \(\lambda_k\) and have \(r(w_k,A_k)\) of the form \(r(w_k,A_k)\equiv [x_kx_k^T]^m(A_kw_k-\lambda_k(w_k)w_k) \) where \(m\in \{0,-1,1\}\). As an intermediate step in analyzing stochastic convergence, the ordinary differential equation \(dw/{dt}=\lim_{k}E[r(w,A_k)]\) is used. Convergence of known methods and of two new modifications are discussed. Numerical examples deal mainly with \(n=10\), \(k=250\), \(k=500.\)
- A survey of conjugate gradient algorithms for solution of extreme eigen-problems of a symmetric matrix
- Analysis of recursive stochastic algorithms
- scientific article; zbMATH DE number 48727 (Why is no real title available?)
- scientific article; zbMATH DE number 51537 (Why is no real title available?)
- scientific article; zbMATH DE number 53271 (Why is no real title available?)
- scientific article; zbMATH DE number 3451330 (Why is no real title available?)
- scientific article; zbMATH DE number 3892457 (Why is no real title available?)
- On stochastic approximation of the eigenvectors and eigenvalues of the expectation of a random matrix
- Stochastic approximation methods for constrained and unconstrained systems
- On stochastic approximation of the eigenvectors and eigenvalues of the expectation of a random matrix
- Stability and chaos of LMSER PCA learning algorithm
- Convergence analysis of Chauvin's PCA learning algorithm with a constant learning rate
- Global convergence of a PCA learning algorithm with a constant learning rate
- scientific article; zbMATH DE number 3880066 (Why is no real title available?)
- Linear Prediction Based Blind Source Extraction Algorithms in Practical Applications
- Efficient coordinate-wise leading eigenvector computation
- Adaptive multiple minor directions extraction in parallel using a PCA neural network
- Online Principal Component Analysis in High Dimension: Which Algorithm to Choose?
- Adaptive algorithms for solving eigenvalue problems in the variable computer environment of supercomputers
- Convergence proof of matrix dynamics for online linear discriminant analysis
- The modified natural power method for principal component computation
- Global convergence of Oja's PCA learning algorithm with a non-zero-approaching adaptive learning rate
- Fast adaptive algorithms using eigenspace projections
- Adaptive eigenvalue computation: Complexity estimates
This page was built for publication: Adaptive algorithms for first principal eigenvector computation
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q557636)
