Ecole d'ingénieur et centre de recherche en Sciences du numérique

Seminar: Distributed Large-Scale Cholesky Decomposition

Duc Trung NGUYEN - Doctorant

Data Science

Date: 7 décembre 2017

Lieu: Eurecom - Salle de cours 152

Abstract: "We present the design, analysis and evaluation of a distributed algorithm for the factorization of large, dense matrices. Our approach follows the lines of the well-known Cholesky algorithm, and is amenable to a simple, practical and efficient implementation using modern computational models and distributed frameworks; in addition, it does not require specialized hardware nor impose a rigid structure to matrix storage. Our experiments, which we execute using an Apache Spark implementation of our algorithm, indicate that our design can exploit low-level linear algebra primitives to compensate communication costs, and has the potential to scale to matrix sizes that are currently beyond what the literature considers feasible." Bio : NGUYEN Duc Trung is a third-year PhD student supervised by Prof. Pietro Michiardi, whose primary research is focused on scalable algorithms of matrix operations on distributed systems

Distributed Large-Scale Cholesky Decomposition