BlockFL: blockchain-enabled decentralized federated learning and model trading
Federated Learning (FL) is a promising privacy-preserving distributed machine learning paradigm. However, there is only a centralized parameter server to aggregate all the local model updates, which brings the challenges of a single point of failure and server overload, especially in large-scale tra...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2022
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/156495 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Federated Learning (FL) is a promising privacy-preserving distributed machine learning paradigm. However, there is only a centralized parameter server to aggregate all the local model updates, which brings the challenges of a single point of failure and server overload, especially in large-scale training scenarios. To achieve secure, reliable, and scalable FL, we leverage a sharding technique to improve scalability of the Blockchain-based Federated Edge Learning (BFEL) framework with a main chain and multiple subchains in [Kang et al., 2020]. Specifically, to release the cross-chain transaction processing workload of the main chain, the number of working consensus nodes for the main chain can be divided into multiple clusters to process multiple cross-chain transactions in parallel. This method helps reduce the execution time for FL task training and improve transaction throughput on the main chain. This project presents a working prototype to utilize blockchain and sharding techniques, thereby scaling up decentralized FL for secure, scalable and large-scale FL task training. |
|---|