Guangfa Group’s Federated Learning Exploration, Platform Construction, and the Book “Federated Learning Principles and Applications”

The article introduces Guangfa Group’s recent work in privacy‑preserving computation and federated learning, and announces the publication of the monograph “Federated Learning Principles and Applications”, which records the group’s research progress and practical experiences.

It first reviews the policy background—GDPR, CCPA, China’s Data Security Law and Personal Information Protection Law—and explains how stricter data‑use regulations have driven the emergence of privacy‑computing technologies.

The company’s privacy‑computing platform is described as a three‑layer architecture: a big‑data foundation, a blockchain‑based audit layer, and a middle layer that integrates a distributed decision engine, secure multi‑party computation (MPC), and a federated learning platform built on the open‑source FATE framework. Guangfa has contributed custom algorithms to FATE and participates in its technical steering committee.

Among the technical contributions is a secure multi‑party summation algorithm based on verifiable secret sharing. The secret values of each party (e.g., a0, b0, c0 ) are split into polynomial shares, distributed, summed, and reconstructed using Lagrange interpolation, ensuring that no party learns the others’ raw data. Performance tests show linear scaling with data size and participant count, achieving 100% accuracy even with up to one million records.

Four concrete application scenarios are highlighted: (1) joint statistical analysis of customer migration across subsidiaries, leveraging Paillier homomorphic encryption; (2) precise marketing for group‑insurance products using a hybrid local‑model + federated model approach; (3) intelligent risk control in securities, where SecureBoost outperforms local XGBoost and blockchain records audit information; and (4) cross‑domain data verification with custom encryption schemes that provide millisecond‑level response times.

The book’s nine chapters cover the evolution of federated learning, algorithmic foundations, FATE architecture and deployment, end‑to‑end modeling workflows, case studies, and an appendix on cryptographic primitives, offering a practical guide for practitioners.

Finally, the article looks ahead to industry trends: standardization efforts, the need for unified interoperability, deeper integration of blockchain for trustworthy data sharing, and continued performance improvements through hardware acceleration.