Federated learning data governance and specification
Federated learning data governance focus topic reference
|Major issues||Detailed topics (examples)|
|1. The issues of capital, privacy, security and ethics under the federated learning model||1. The federated learning model may solve/derive privacy, personal information protection, and information security issues.|
2. Privacy and information security risk assessment and the formulation of acceptable risks underfederated learning.
3. Relevant technologies such as the de-identification of individual resources and the interface of model database.
4. The consistency of the internal data collection, processing, and utilization management of the participants (or institutions), including: the informed consent of the data subject and the design of the withdrawal mechanism.
5. Application and review focus of research ethics (IRB).
6. Privacy and information security environment requirements for data or model porting.
|2. The model of federated learning, intellectual wealth-related rights and obligations||1. The rights and obligations of the model co-developer, such as: the rights and obligations of model monitoring and subsequent improvement under the continuous accumulation of data; the handling method and legal relationship after the model co-developer exits the cooperation mechanism, etc.|
2. The distribution of intellectual property rights in the output model.
|3. ” New data governance” models||1. Participants establish a federated learning database to provide governance for external AI models to enter retraining or verification, such as: application, review, use of feedback management procedures, rights and obligations, etc.|
2. Management requirements for the roles of data providers, project PI, IRB, service platforms, etc.
3. New algorithm or technical requirements required by the alliance, such as: initial model quality management, federated learning privacy protection, model aggregation algorithm, cross-level federated learning algorithm, participant contribution tracking technology, etc.
4. The design of standardized processes, management specifications, forms required by the alliance and integrated into the platform mechanism(ethic in design/governance in design)。
|4. Experiments or effectiveness of output models under federated learning||1. The relevant criteria of AI in clinical trials or field trials.|
2. How to improve the credibility of the model, such as: establishing a flexible, efficient and friendly clinical verification plan, integrating domestic and foreign SaMD inspection mechanisms or AI software certification, etc.