Bridging Data Silos in Finance via Federated Learning

Abstract

The financial industry faces the challenge of balancing data utilization and privacy protection. Federated learning (FL) offers a promising solution by enabling secure collaborative training. This paper focuses on an analysis of the key technologies for several financial applications that can benefit from FL. Specifically, we examine precision marketing based on multimodal FL (MMFL), anti-money laundering strategies leveraging federated graph learning (FGL), and credit card risk assessment utilizing vertical federated learning (VFL). Furthermore, we identify the key challenges in large-scale applications of FL in the financial industry. Additionally, we propose forward-thinking applications of FL in the finance sector, including the use of federated large language models (LLMs) for intelligent customer service (ICS) and decentralized FL integrated with blockchain for financial audit. Finally, we conduct a case study by using a consumer complaints dataset to verify the feasibility and effectiveness of federated LLMs in ICS.