Responsible and Effective Federated Learning in Financial Services: A Comprehensive Survey

The financial sector is increasingly leveraging Artificial Intelligence (AI) to deliver intelligent, automated, and personalized services. However, it encounters significant data privacy challenges due to the dispersion of financial data across various entities. Federated Learning (FL) offers a potential solution by facilitating AI model training at the source of data, albeit with certain challenges. Irresponsible utilization of FL can compromise stakeholder interests, and the prevalent heterogeneity in data spaces in numerous financial FL scenarios can impede FL's performance. These complications necessitate the development of a Responsible and Effective Federated Learning (RE-FL) system in finance. In this paper, we explore the interdisciplinary field of RE-FL in finance and guide readers to understand this area thoroughly. We present a taxonomy of RE-FL approaches that address the concerns of stakeholders in FL-based financial services and identify six major dimensions: accountability, controllability, fairness, privacy, security, and effectiveness. We also propose potential directions for future research. To our understanding, this is the first literature review conducted on RE-FL in the financial sector.