1: \begin{abstract}

2: %We study the high-dimensional median as robust aggregation for distributed synchronous Stochastic Gradient Descent~(SGD). To be more specific, we study parameter server architecture, in which the workers have  arbitrary~(Byzantine) failures. We propose two Byzantine-resilient gradient aggregation rules with provable convergence. Furthermore, using marginal median, we propose the first dimensionally Byzantine-resilient algorithm for distributed synchronous SGD. Empirical results also show good performance. 

3: 

4: We propose three new robust aggregation rules for distributed synchronous Stochastic Gradient Descent~(SGD) under a general Byzantine failure model. The attackers can arbitrarily manipulate the data transferred between the servers and the workers in the parameter server~(PS) architecture. We prove the Byzantine resilience properties of these aggregation rules. Empirical analysis shows that the proposed techniques outperform current approaches for realistic use cases and Byzantine attack scenarios.

5: \end{abstract}

6: