Robustness of Iteratively Pre-Conditioned Gradient-Descent Method: The Case of Distributed Linear Regression Problem

This letter considers the problem of multi-agent distributed linear regression in presence system noises. In this problem, comprises multiple agents wherein each agent locally observes a set data points, and agents' goal is to compute model that best fits collective points observed by all agents. We consider server-based architecture where interact with common server solve problem; however, cannot access points. practical scenario either has observation noise, i.e., are corrupted, or process computations performed corrupted. noise-free systems, recently proposed algorithm, named Iteratively Pre-conditioned Gradient-descent (IPG) method, been claimed converge faster than related methods. letter, we study robustness IPG against both noise noise. empirically show method compares favorably state-of-the-art algorithms.