Enhancing Robustness of Federated Learning via Server Learning

Forum topic · 小凯 · 2026-04-06

Summary

A new arXiv paper (2604.03226) by Van Sy Mai, Kushal Chakrabarti, Richard J. La, and colleagues explores using server learning to strengthen the robustness of federated learning against malicious attacks, even when client training data are non-IID. The authors propose a heuristic algorithm that combines server-side learning with client update filtering and geometric median aggregation. Experiments show significant improvements in model accuracy even when the fraction of malicious clients is very high—exceeding 50% in some cases. Notably, the server's dataset can be small and even synthetic, and its distribution need not closely match the aggregated data distribution of the clients, making the approach practical for real-world deployments where server-side data resources are limited. The paper was published on April 3, 2026, in the machine learning field.

Paper Overview

Field: Machine Learning (ML)
Authors: Van Sy Mai, Kushal Chakrabarti, Richard J. La, et al.
Published: 2026-04-03
arXiv: 2604.03226

Abstract

This paper explores the use of server learning for enhancing the robustness of federated learning against malicious attacks even when clients' training data are not independent and identically distributed. The authors propose a heuristic algorithm that uses server learning and client update filtering in combination with geometric median aggregation.

Key Findings

The proposed approach combines three mechanisms:
Server learning (training on a server-side dataset)
Client update filtering
Geometric median aggregation of client updates
Experiments demonstrate significant improvement in model accuracy even when the fraction of malicious clients is high — exceeding 50% in some cases.
The server's dataset can be small and even synthetic, and its distribution does not need to closely match the aggregated distribution of clients' data.
Robustness is maintained even under non-IID client training data, a common challenge in practical federated learning deployments.

Original Abstract

> This paper explores the use of server learning for enhancing the robustness of federated learning against malicious attacks even when clients' training data are not independent and identically distributed. We propose a heuristic algorithm that uses server learning and client update filtering in combination with geometric median aggregation. We demonstrate via experiments that this approach can achieve significant improvement in model accuracy even when the fraction of malicious clients is high, even more than $50\%$ in some cases, and the dataset utilized by the server is small and could be synthetic with its distribution not necessarily close to that of the clients' aggregated data.

Full paper: arXiv:2604.03226