Privacy-preserving federated learning framework based on Hyperledger Fabric and Hyperledger Aries.
Description
This project utilizes Hyperledger Fabric to develop a federated learning framework with committee consensus. Hyperledger Fabric is used for storing and tracing global model update exchange. A customized committee consensus mechanism is required to reduce a load of consensus computation by selecting a random smaller subset of nodes to participate in consensus each round. In order to protect the identity of local data owners, the proposed framework will be integrated with Indy, Aries, and Ursa stack projects to offer self-sovereign identity. The mentee will work with the project mentors to gather and validate the requirements, and design the appropriate technical solution.
Additional Information
https://github.com/OpenMined/PyAriesFL
https://github.com/blockchain-systems/ScaleSFL
Peng, Zhe, Jianliang Xu, Xiaowen Chu, Shang Gao, Yuan Yao, Rong Gu, and Yuzhe Tang. "VFChain: enabling verifiable and auditable federated learning via blockchain systems." IEEE Transactions on Network Science and Engineering 9, no. 1 (2021): 173-186.
Qi, Yuanhang, M. Shamim Hossain, Jiangtian Nie, and Xuandi Li. "Privacy-preserving blockchain-based federated learning for traffic flow prediction." Future Generation Computer Systems .
Learning Objectives
- The mentee will be trained to work with various technologies such as blockchain, virtualization, digital credentials, and secrets management incorporated into Hyperledger Fabric and Hyperledger Aries and Indy.
- The mentee will gain a well-rounded experience in developing a blockchain-based federated learning approach and contribute to the research work.
- The mentee will also be introduced to open source culture and collaborative team working.
- The mentee will get hands-on experiences with software development best practices.
- The mentee will have the opportunity to practice Agile development methodology.
Expected Outcome
- Open source implementation, testing and documentation for federated learning framework.
- Academic paper that discuss the details of system design and implementation.
Relation to Hyperledger
Hyperledger Fabric, Aries, and Indy.
Education Level
Graduate students (Master's or PhD) are preferred, but the experienced undergraduate students will be considered.
Skills
Knowledge of Hyperledger Fabric, Aries, and Indy, and one of the federated learning frameworks, such as Flower or PySyft.
Future plans
The future work focuses on enhancing consensus mechanisms to tackle the potential performance and scalability issues.
Contributions such as adding additional features, improving the documentation, reporting issues/bugs and providing feedback from the community and interested parties are highly encouraged.
Preferred Hours and Length of Internship
Full time preferred, but part-time also possible.
Mentor(s) Names and Contact Info
Sara Rouhani, Assistant Professor, University of Manitoba sara.rouhani@umanitoba.ca
Vahid Pourheidari, Solutions Architect, ISM Canada vahid.pourheidari@gmail.com
Rui Pan, Blockchain developer, Rewatt Power and Grain Discovery rui@rewattpower.com