On Distributed Solution for Simultaneous Linear Symmetric Systems

Cholesky Decomposition is the primary approach which is used to solve Symmetric and Positive Definite (SPD) systems but is inherently iterative making it very difficult to parallelize as calculations at each partition require elements from other partitions. In this paper, we present two distributed...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2020 IEEE International Conference on Big Data (Big Data) S. 5780 - 5782
Hauptverfasser: Misra, Chandan, Parasrampuria, Utkarsh, Bhattacharya, Sourangshu, Ghosh, Soumya K.
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 10.12.2020
Schlagworte:
Apache Spark
Big Data
Cholesky Decomposition
Distributed Algorithm
Iterative methods
Linear Algebra
Matrix decomposition
Matrix Inversion
Partitioning algorithms
Scalability
Sparks
Symmetric matrices
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Cholesky Decomposition is the primary approach which is used to solve Symmetric and Positive Definite (SPD) systems but is inherently iterative making it very difficult to parallelize as calculations at each partition require elements from other partitions. In this paper, we present two distributed block-recursive approaches to solve large SPD systems - the symmetric version of the state-of-the-art Strassen's algorithm and Cholesky based inversion algorithm. We show experimentally that both the approaches have good scalability and Cholesky based approach is more efficient as it uses fewer matrix multiplications in each recursion level than Strassen based algorithm.
DOI:10.1109/BigData50022.2020.9377840