A Shapley-Based Robust Federated Learning Framework with Data Distribution Weighting Constraints

View at publisher Summarise with VIRAT
Authors: Weiqing Zhang, Jianhua Liu
Cited by: 0
References: 18

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Federated learning constructs global models using the 'local training and parameter aggregation' approach. However, in scenarios involving severe non-independent and identically distributed (non-IID) data, label noise, and malicious clients, existing aggregation methods suffer from training instability and imbalanced client incentive mechanisms. This can result in model updates that deviate from the global optimum. To address this issue, we present the Data-distribution Weight-Constrained Shapley (DWS) federated learning method. Based on Shapley value contribution estimation, we method use the Wasserstein distance to quantify data distribution heterogeneity and constrain Shapley value bias. This fusion yields a unified DWS score, which is used to generate client sampling probabilities and aggregation weight coefficients. Experimental validation on rail defect detection and MNIST classification demonstrates that the proposed method achieves significant improvements over traditional Shapley valuebased federated learning algorithms in terms of convergence speed, final accuracy, precision, and recall.

Fields of Study & Subjects

Privacy-Preserving Technologies in Data Machine Learning and Data Classification Stochastic Gradient Optimization Techniques

Publication details

Journal / source
International Conference on Electrical Automation and Artificial Intelligence (ICEAAI)
Pages
1278–1283
Publication date
16 January 2026
Language
EN