Interactive Byzantine-Resilient Gradient Coding for General Data Assignments

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Jain, Shreyas, Maßny, Luis, Hofmeister, Christoph, Yaakobi, Eitan, Bitar, Rawad
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911767686283264
author Jain, Shreyas
Maßny, Luis
Hofmeister, Christoph
Yaakobi, Eitan
Bitar, Rawad
author_facet Jain, Shreyas
Maßny, Luis
Hofmeister, Christoph
Yaakobi, Eitan
Bitar, Rawad
contents We tackle the problem of Byzantine errors in distributed gradient descent within the Byzantine-resilient gradient coding framework. Our proposed solution can recover the exact full gradient in the presence of $s$ malicious workers with a data replication factor of only $s+1$. It generalizes previous solutions to any data assignment scheme that has a regular replication over all data samples. The scheme detects malicious workers through additional interactive communication and a small number of local computations at the main node, leveraging group-wise comparisons between workers with a provably optimal grouping strategy. The scheme requires at most $s$ interactive rounds that incur a total communication cost logarithmic in the number of data samples.
format Preprint
id arxiv_https___arxiv_org_abs_2401_16915
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Interactive Byzantine-Resilient Gradient Coding for General Data Assignments
Jain, Shreyas
Maßny, Luis
Hofmeister, Christoph
Yaakobi, Eitan
Bitar, Rawad
Information Theory
Distributed, Parallel, and Cluster Computing
We tackle the problem of Byzantine errors in distributed gradient descent within the Byzantine-resilient gradient coding framework. Our proposed solution can recover the exact full gradient in the presence of $s$ malicious workers with a data replication factor of only $s+1$. It generalizes previous solutions to any data assignment scheme that has a regular replication over all data samples. The scheme detects malicious workers through additional interactive communication and a small number of local computations at the main node, leveraging group-wise comparisons between workers with a provably optimal grouping strategy. The scheme requires at most $s$ interactive rounds that incur a total communication cost logarithmic in the number of data samples.
title Interactive Byzantine-Resilient Gradient Coding for General Data Assignments
topic Information Theory
Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2401.16915