Federated Learning for Financial Data Sharing: Compliance, Auditable Solutions, and Blockchain‑Enabled Marketplace

Speaker Xiang Xiaojia, Deputy General Manager of Guangda Technology, introduces the regulatory landscape and technical challenges of implementing federated learning for data sharing within a financial holding group, emphasizing the need for compliance and business value.

The latest regulations from the People’s Bank of China impose strict protection of personal financial information, adding to a broader multi‑head governance framework that includes state secrets, commercial secrets, antitrust, and cybersecurity laws.

To meet these requirements, the team proposes an auditable federated learning approach that adds a blockchain‑based immutable evidence layer after the "last mile" of federated computation, enabling institutions to provide auditors with verifiable records without exposing raw data.

A concrete PSI (Private Set Intersection) example uses RSA‑based hashing and public‑key encryption to satisfy the Securities Regulatory Commission’s Article 34 (Regulation No. 152), ensuring that original data never leaves the securities‑fund institution while still allowing aggregated statistics.

Three solution tiers are described: aggregate‑level models, PSI‑based collaborative models, and fine‑grained regional models, each accompanied by an incentive mechanism that records data contributions and rewards participants with tokenized points.

The "Biyan" paid‑data‑sharing sandbox combines a blockchain‑driven token system (stable points EBP and securities token EBST) with a federated engine (FATE) to price, settle, and audit data assets, allowing secure, on‑chain data exchange without violating regulations.

Future outlook highlights the need for mature, out‑of‑the‑box federated learning toolchains, standardized cross‑platform collaboration, financial‑grade security algorithms, audit tools, deeper blockchain integration, and an expanding catalog of data assets on the federated marketplace.