Hong Kong Quality Assurance Agency (HKQAA)
Although Federated learning (FL) has begun to help various industries as a booming AI technique because of its privacy-preserving motivation, this technique cannot still wholly dispel peoples’ concerns about privacy leakage. One of the main reasons is that current FL commercial solutions are still “black box” to most of the common industry users [1]. They are reluctant to accept this technique easily due to the lack of technical references and a standard prototype. The other reason is the automated objective evaluation approach is also still absent in the limited available AI assessments. Finally, the data privacy regulatory authority also has such requirement that organisations need to adopt reliable, robust, and secure/human oversight AI systems [2]. This project aims to design and implement an effective application integration software platform that validates the federated learning capabilities of AI commercial solutions. The objective of the formation of this software platform is to help enterprises to evaluate the basic function, model performance, and system security of their FL products in the automatic and standard approaches. This platform will be beneficial for a wide variety of stakeholders that include FL technical solution vendors, industry data owners, audit service agencies, data privacy regulatory authority, and the public. [Reference] [1] Kairouz, Peter, et al. "Advances and open problems in federated learning." Foundations and Trends® in Machine Learning 14.1–2 (2021): 1-210. [2] https://www.pcpd.org.hk//english/resources_centre/publications/files/guidance_ethical_e.pdf.