A1310
Title: Robust personalized federated learning with sparse penalization
Authors: Xiaofei Zhang - Zhongnan University of Economics and Law (China) [presenting]
Abstract: Thanks to its advantage in collaborative learning with distributed data, Federated learning (FL) is an emerging topic. Due to the local data-generating mechanism heterogeneity, it is important to consider personalization when developing federated learning methods. A personalized federated learning (PFL) method for addressing the robust regression problem is proposed. Specifically, the aim is to learn the regression weight by solving a Huber loss with the sparse fused penalty. Additionally, the personalized federated learning for robust and sparse regression (PerFL-RSR) algorithm was designed to solve the estimation problem in the federated system efficiently. Theoretically, the convergence property of the proposed PerFL-RSR algorithm is shown, and then the proposed estimator is shown to be statistically consistent. Thorough experiments and real data analysis are conducted to corroborate the theoretical results of the proposed personalized federated learning method.
