Federated Learning for Financial Data Sharing: Compliance, Incentives, and Technical Innovations

Speaker Xiang Xiaojia, Deputy General Manager of Guangda Technology, introduces the regulatory background for financial data sharing, highlighting recent People's Bank of China regulations that tighten protection of personal and financial information.

To comply with strict regulations, the team adopts technical measures such as federated learning and multi‑party secure computation, adding audit capabilities that generate immutable evidence (e.g., via blockchain) for downstream auditors.

A case study shows how a securities fund institution was penalized despite encrypting shared data, underscoring the necessity of compliant solutions.

The proposed audit‑enabled federated learning framework adds a post‑processing layer that records immutable proof of data usage, enabling institutions to demonstrate compliance while preserving data privacy.

Three solution tiers are described: aggregate customer scoring models, PSI‑style collaborative models across institutions, and regional models with finer granularity, all leaving audit trails.

To overcome business reluctance, an incentive‑driven data‑sharing sandbox (“Bì'àn”) is introduced, leveraging blockchain‑based tokens to reward data contributors and ensure fair, auditable transactions.

The sandbox supports raw data, data products, and model assets, all kept within domain boundaries, priced with a dual‑track token system (stable and incentive tokens), and settled on a distributed ledger (Quorum) using the FATE federated learning engine.

Technical innovations include a Federated AI Hub and front‑end federated inference, which shift inference to client devices, reduce reliance on central servers, and improve performance.

The speaker concludes with future expectations: more mature federated learning toolchains, standardized cross‑platform collaboration, stronger financial‑grade security, additional compliance tools, deeper blockchain integration, and expanded data asset listings on the federated platform.