Current browse context:
cs.PF
Change to browse by:
References & Citations
Computer Science > Performance
Title: Eigenvector Component Calculation Speedup over NumPy for High-Performance Computing
(Submitted on 12 Feb 2020 (v1), revised 14 Feb 2020 (this version, v2), latest version 16 Jun 2020 (v4))
Abstract: Applications related to artificial intelligence, machine learning, and system identification simulations essentially use eigenvectors. Calculating eigenvectors for very large matrices using conventional methods is compute-intensive and renders the applications slow. Recently, Eigenvector-Eigenvalue Identity formula promising significant speedup was identified. We study the algorithmic implementation of the formula against the existing state-of-the-art algorithms and their implementations to evaluate the performance gains. We provide a first of its kind systematic study of the implementation of the formula. We demonstrate further improvements using high-performance computing concepts over native NumPy eigenvector implementation which uses LAPACK and BLAS.
Submission history
From: Shrey Dabhi [view email][v1] Wed, 12 Feb 2020 13:44:41 GMT (890kb)
[v2] Fri, 14 Feb 2020 16:40:42 GMT (186kb,D)
[v3] Mon, 13 Apr 2020 20:27:06 GMT (339kb,D)
[v4] Tue, 16 Jun 2020 07:21:06 GMT (339kb,D)
Link back to: arXiv, form interface, contact.