A simplified convergence theory for Byzantine resilient stochastic gradient descent

In distributed learning, a central server trains model according to updates provided by nodes holding local data samples. the presence of one or more malicious servers sending incorrect information (a Byzantine adversary), standard algorithms for training such as stochastic gradient descent (SGD) fail converge. this paper, we present simplified convergence theory generic Resilient SGD method originally proposed Blanchard et al. (2017) [3]. Compared existing analysis, shown stationary point in expectation under assumptions on (possibly nonconvex) objective function and flexible gradients.