May 2020
In this Ekkono SWP (Short White Paper) we discuss the Pros and Cons with Federated Learning. Machine Learning for IoT has traditionally been done by uploading all data from each connected device to the cloud to train a generic model that can be distributed and applied to all devices. The advantage of this type of centralized learning is that the model can generalize based on data from a group of devices and thus instantly work with other compatible devices. Centralized learning also entails that data can explain all variations in the devices and their environment.
The downside of individualized decentralized learning is that the learning cannot generalize from results derived on other devices as in centralized learning. Combining centralized and individualized decentralized learning is a complex and intricate subject, requiring different solutions for different ML techniques. A new research field, Federated Learning, is addressing this issue. In essence, Federated Learning is a machine learning technique to train algorithms across decentralized edge devices while holding data samples locally.
Ekkono's SWP are documented results of our #openfika webinars.
The webinars are also recorded and available to watch in our YouTube channel or below: