Design Considerations and Guidelines
under the orchestration of a central server ( e . g . service provider ), while keeping the training data decentralized .”’
• OEM – Original Equipment Manufacturer
Due to growing competition and data privacy concerns many organizations are reluctant to share their data with each other or on cloud infrastructures ( for data pooling ); and thus , are deprived access to the variety and veracity of having data gathered from multiple sources to train ML models ( Mohr , 2021 ). This also hinders the potential to unlock value from unused datasets . FedL enables organizations to mutually benefit from each other ' s data by collaboratively training robust ML models without having to share their raw data ( Kairouz , 2021 ).
A FedL setup typically consists of several iterative phases to support model training which is initialized by FedL clients downloading a common model from a trusted centralized server . Clients proceed to train the model using data collected locally . Once the model is trained , the client shares only the updated model parameters with the trusted server . This is followed by the aggregation of the received model updates ( from all clients ) to create an updated global model that can be downloaded by the clients for the next iteration of training . The process terminates once the clients reach a consensus regarding the optimally of the global model . This approach allows the model to be exposed to a significantly larger pool of data that would be impossible for a single organization to possess alone .
Figure 2-1 : Ecosystems and Smart-Manufacturing Use Cases for Participants
The first step towards implementing FedL in smart manufacturing scenarios is to identify an appropriate use case for FedL , i . e ., where collaborative model training and / or sharing can
IIC Journal of Innovation 21