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