Current browse context:
cs.IT
Change to browse by:
References & Citations
Computer Science > Information Theory
Title: Speeding Up Private Distributed Matrix Multiplication via Bivariate Polynomial Codes
(Submitted on 16 Feb 2021 (v1), last revised 13 Jul 2021 (this version, v2))
Abstract: We consider the problem of private distributed matrix multiplication under limited resources. Coded computation has been shown to be an effective solution in distributed matrix multiplication, both providing privacy against the workers and boosting the computation speed by efficiently mitigating stragglers. In this work, we propose the use of recently-introduced bivariate polynomial codes to further speed up private distributed matrix multiplication by exploiting the partial work done by the stragglers rather than completely ignoring them. We show that the proposed approach reduces the average computation time of private distributed matrix multiplication compared to its competitors in the literature while improving the upload communication cost and the workers' storage efficiency.
Submission history
From: Burak Hasircioglu [view email][v1] Tue, 16 Feb 2021 17:40:30 GMT (86kb)
[v2] Tue, 13 Jul 2021 17:56:56 GMT (86kb)
Link back to: arXiv, form interface, contact.