We report on \emph{Krum}, the first \emph{provably} Byzantine-tolerant aggregation rule for distributed Stochastic Gradient Descent (SGD). Krum guarantees the convergence of SGD even in a distributed setting where (asymptotically) up to half of the workers can be malicious adversaries trying to attack the learning system.
Details
Title
Brief Announcement: Byzantine-Tolerant Machine Learning
Note
This work has been supported in part by the European ERC Grant 339539 - AOC and by the Swiss National Science Foundation under the grant 200021\_169588 TARBDA (a Theoretical Approach to Robustness in Biological Distributed Algorithms)