Stochastic gradient descent for streaming linear and rectified linear systems with Massart noise

arXiv:2403.01204v1 Announce Type: new
Abstract: We propose SGD-exp, a stochastic gradient descent approach for linear and ReLU regressions under Massart noise (adversarial semi-random corruption model) for the fully streaming setting. We show novel nearly linear convergence guarantees of SGD-exp to the true parameter with up to $50\%$ Massart corruption rate, and with any corruption rate in the case of symmetric oblivious corruptions. This is the first convergence guarantee result for robust ReLU regression in the streaming setting, and it shows …

abstract adversarial arxiv convergence corruption cs.lg cs.na gradient linear math.na noise novel random relu show stat.ml stochastic streaming systems true type