A1072
Title: Byzantine-robust distributed learning under heterogeneity via convex hull search
Authors: Zhao Chen - Fudan University (China) [presenting]
Abstract: In modern massive data modelling, distributed learning plays a critical role in enhancing scalability, efficiency and privacy protection. The heterogeneity and robustness of a distributed learning algorithm are key aspects related to the accuracy and reliability of learning results. Under the common framework of statistical learning, the convex hull search algorithm is proposed, which has four main advantages: fast convergence, high accuracy, adjustable robustness, and tuning friendliness. The corresponding convergence and asymptotic normality result for the CHS algorithm are established, which shows its adaptability to data heterogeneity. The algorithm for regression and clustering tasks is exemplified through synthetic data. Furthermore, real energy consumption data is implemented for Gaussian process regression hyperparameters optimization. Existing numerical results confirm the superiority and exhibit the wide applicability of the algorithm
