Variance Reduction-Principal Component Analysis

Brief Description

The research aim is to extend the exponential convergence rate algorithm, developed by Shamir [1]. In particular, the algorithm is able to find the first eigenvector of the data matrix with exponential rate. In this research, we aim to develop an algorithm for finding the multiple eigenvetors of a matrix simultaneously with the same convergence rate.

The developed algorithm is promising, although consistent convergence is not achieved. Hence, the further research is required. More details on the algorithm and significant results is provided in the report below.

Contributors

Ziyu Wang & Tan Bui-Thanh

Report

Variance Reduction-Principle Component Analysis

Reference

[1] Shamir, O. A Stochastic PCA and SVD Algorithm with an Exponential Convergence Rate.arXiv:1409.2848, 2015