2.4.4. DLT-based FL enabler
2.4.4.1. Introduction
This enabler will foster the use of DLT-related components to exchange the local, on-device models (or model gradients) in a decentralised way.
2.4.4.2. Features
The DLT can act as a component to manage AI contex-tual information and prevent any alteration to the data. The alteration of data is a threat to the Fed-erated Learning approach and the DLT can help in mitigating the threat. Moreover, the enabler will allow mitigating single-point of failures. Finally, the enabler can be charged with validating the in-dividually trained models to rule out malicious updates that can harm the global model.
2.4.4.3. Place in architecture
The DLT-based FL enabler is part of the vertical plane DLT enablers.
2.4.4.4. User guide
The user guide will be determined after the release of the enabler.
2.4.4.5. Prerequisites
Hyperledger Fabric 2.2, Hyperledger Fabric CA 1.4
2.4.4.6. Installation
The installation procedure is under development.
2.4.4.7. Configuration options
The enabler is prepared to run in a K8S environment. The creation is prepared to be autonomous in such a working environment. The service consumer will be required to communicate with the server using the described Rest interface.
2.4.4.8. Developer guide
The DLT-based FL enabler is build using Hyperledger Fabric Framework. Smart contracts are written in Go.
2.4.4.9. Version control and release
Gitlab will be used as a version control and release tool.
2.4.4.10. License
Will be determined after the release of the enabler.
2.4.4.11. Notice(dependencies)
Dependency list and licensing information will be provided