Research on the Application of Security- Weighted Aggregation in Federated Learning for Financial Risk Control

Abstract

Federated learning is a privacy-preserving distributed machine learning approach that enables collaborative training by keeping data local and sharing only model parameters. This effectively addresses the challenges of “data silos” and privacy regulations in the field of financial risk control. This paper systematically elaborates on the concepts, development history, architecture, and core mechanisms of federated learning, with a focus on analyzing the principles and implementation of securityweighted aggregation technology, including its reliance on key technologies such as homomorphic encryption, differential privacy, and secure multi-party computation. Besides, this study examines how this technology applies to fintech, outlines common practices, and highlights challenges in technical performance, cross-platform collaboration, and compliance, with possible solutions. The results show that federated learning, using secure aggregation, enables the financial industry to share data and conduct joint risk modeling while staying compliant.