Privacy Computing: Technical Routes Overview and Ant Group’s Contributions

Introduction: Privacy computing is a key technology for future data element markets, with many complex technical routes that must be selected based on specific business scenarios.

Technical route comparison: The article evaluates four typical privacy computing approaches—classic MPC, federated learning, TEE, and MPC proxy—using a color‑coded rating (green = advantage, yellow = moderate, red = disadvantage) across ten dimensions: security, application development cost, user operation cost, calculation accuracy, calculation performance, number of data participants, data control, proprietary hardware cost, trust root & autonomy, and overall suitability.

Key findings include: MPC offers provable security but high development cost and limited scalability; federated learning provides better performance for machine‑learning workloads but weaker provable security; TEE delivers native performance and accuracy but depends on hardware trust; proxy MPC can improve scalability with multiple agents but suffers from collusion risk.

Ant Group’s contributions: The company has built the “隐语” (YinYu) privacy‑computing framework, which integrates resource management, mixed clear‑ciphertext scheduling, AI/BI privacy algorithms, and a visual UI, aiming for modular and layered deployment. It also promotes privacy‑computing applications across marketing, credit risk, security, insurance, healthcare, and government, and actively participates in standards development, contributing to dozens of white papers and over 40 standards in national and industry bodies.

Conclusion: No single technology solves all challenges; selection must consider security assumptions, performance, functional requirements, and hardware availability to choose the most appropriate privacy‑computing solution.