Design Considerations and Guidelines
As the original data can be re-constructed from shared model updates ( Kairouz , 2021 ) there is a requirement to further protect sensitive manufacturing raw data against the attackers . Traditionally , trusted aggregation mechanisms are employed to deal with this issue . However , it becomes a major challenge in cross-silo FedL across multiple manufacturing organizations .
Due to the preferred fully distributed control in such scenarios , a consensus relating to the global model can be achieved by utilizing Distributed Ledger Technologies - DLTs ( Isaja , 2018 ) such as Blockchain . A corner stone to any FedL process is data integrity and DLT provides intrinsic properties that can ensure data integrity along a value chain and as such provides a single source of truth that can be used to build reliable models and analysis .
Type of Clients Orchestration Mode of operation Technologies
IoT Edge devices |
Edge servers |
Cloud |
|
|
servers |
Centralized |
Hybrid |
|
Decentralized |
Crossdevice |
Cross-silo |
✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
✓ ✓ ✓
5G , IIoT , Edge Computing , Light-weight machine learning , Digital twin
✓ |
✓ |
✓ |
Deep learning , Differential |
✓ |
✓ |
✓ |
privacy ,
Homomorphic
|
✓ |
✓ |
✓ |
encryption |
✓ ✓ ✓
Big data , Deep learning , Homomorphic encryption , Distributed ledger technologies
Table 4-1 : Possible Configurations for FedL Implementation and Required Technological Drivers
Many platforms are quickly emerging that provide a marketplace for data and model sharing ( Open Application Network , dataspace , IoTA ). These marketplaces leverage Blockchain to provide a mechanism for producers and consumers of datasets that can leverage vast quantities of data in a collaborative , fair and transparent manner . DLT offers the opportunity to monetize data exchange thus incentivizing organizations to collaborate and share model updates .
IIC Journal of Innovation 29