Straggler-Tolerant Iterative Methods for Linear Systems and Eigenvector Computations with Partial Matrix-Vector Products
Linear Algebra seminar by Vasileios Kalantzis, IBM Research
Hosted by Institute for Computational and Experimental Research in Mathematics (ICERM), Brown University
Tuesday 16:00 New York (GMT-4)
Recording available
Abstract
Vasileios Kalantzis develops iterative linear algebra algorithms that tolerate incomplete matrix-vector products in controller-worker cloud systems. Richardson and Chebyshev schemes solve linear systems using randomly available product entries, replacing missing entries with zero. For dominant eigenvectors, modified power iterations substitute zeros, previous entries, or averages of partial iterates for delayed components. The talk presents convergence results in expectation and numerical experiments on sparse matrices for both problem classes.
Topics
Related seminars
Asynchronous Iterative Methods: From Numerical Solvers to Reinforcement Learning
Related research
Structured Matrix Learning from Matrix-Vector Products
Related research
Fault-Tolerant, Distributed In-Memory Computing for Large-Scale Linear Algebra and Optimization: An Algorithm–Hardware Co-Design Approach
Related research