The Hidden Vulnerability of Distributed Learning in Byzantium

While machine learning is going through an era of celebrated success, concerns have been raised about the vulnerability of its backbone: stochastic gradient descent (SGD). Recent approaches have been proposed to ensure the robustness of distributed SGD against adversarial (Byzantine) workers sending \emph{poisoned} gradients during the training phase. Some of these approaches have been proven \emph{Byzantine--resilient}: they ensure the \emph{convergence} of SGD despite the presence of a minority of adversarial workers. We show in this paper that \emph{convergence is not enough}. In high dimension $d \gg 1$, an adver\-sary can build on the loss function's non--convexity to make SGD converge to \emph{ineffective} models. More precisely, we bring to light that existing Byzantine--resilient schemes leave a \emph{margin of poisoning} of $\bigOmega\left(f(d)\right)$, where $f(d)$ increases at least like $\sqrt{d}$. Based on this \emph{leeway}, we build a simple attack, and experimentally show its strong to utmost effectivity on CIFAR--10 and MNIST. We introduce \emph{Bulyan}, and prove it significantly reduces the attacker's leeway to a narrow $\bigO\,( \sfrac{1}{\sqrt{d~}})$ bound. We empirically show that Bulyan does not suffer the fragility of existing aggregation rules and, at a reasonable cost in terms of required batch size, achieves convergence \emph{as if} only non--Byzantine gradients had been used to update the model.

Presented at:
International Conference on Machine Learning, Stockholm, Sweden, July 10-15, 2018
camera ready version available also on ICML proceedings (open access)

Note: The file is under embargo until: 2018-07-16

 Record created 2018-07-17, last modified 2018-09-13

Download fulltext

Rate this document:

Rate this document:
(Not yet reviewed)