Title :
FRPCA: Fast Robust Principal Component Analysis for online observations
Author :
Abdel-Hakim, A.E. ; El-Saban, Motaz
Author_Institution :
Electr. Eng. Dept., Assiut Univ., Assiut, Egypt
Abstract :
While the performance of Robust Principal Component Analysis (RPCA), in terms of the recovered low-rank matrices, is quite satisfactory to many applications, the time efficiency is not, especially for scalable data. We propose to solve this problem using a novel fast incremental RPCA (FRPCA) approach. The low rank matrices of the incrementally-observed data are estimated using a convex optimization model that exploits information obtained from the preestimated low-rank matrices of the original observations. The evaluation results supports the potential of FRPCA for fast, yet accurate, recovery of the low-rank matrices. The proposed FRPCA boosts the efficiency of the traditional RPCA by multiple hundreds of times, while scarifying less than 1% of accuracy.
Keywords :
convex programming; matrix algebra; principal component analysis; FRPCA; RPCA; convex optimization model; fast robust principal component analysis; low-rank matrices; novel fast incremental RPCA approach; online observations; Accuracy; Complexity theory; Convex functions; Data models; Mathematical model; Principal component analysis; Robustness;
Conference_Titel :
Pattern Recognition (ICPR), 2012 21st International Conference on
Conference_Location :
Tsukuba
Print_ISBN :
978-1-4673-2216-4