Federated learning
Federated learning (FL) relies on the idea of decentralizing learning, data analysis, and inference, therefore allowing the user data to be kept within individual devices or local databases (Kaissis et al., 2020; Yang et al., 2019). Through FL, we can benefit from the data of local devices and users, which cannot be stored in a centralized data storage system, to train and improve our machine learning models. As shown in Figure 16.2, a local device or user can provide local data to update the global model and the model we are training and improve the central server. The global model then gets updated and improved and provides updated inferences to the local users and devices.
Figure 16.2 – Schematic representation of updating a model using local data and feeding the global model back to the local devices and users
There are several Python libraries you can benefit from in implementing FL, such as the following:
PySyft...
