Federated Learning is a machine learning technique to train a global model from different clients or devices while keeping the data decentralized. A central server is in charge of coordinating the transfer of the models’ data between the server and the clients, to update the global model while keeping the data private without leaving the device.
How is this connected to edge machine learning? Common use cases tackle the problem of incrementally learning from data directly on the device. A personalized solution is created that works very accurately for that specific device. What if we have many devices of the same type, that are operating under different or similar conditions? Is there a way to make each model smarter, more general, without losing its uniqueness, and being private-aware? That is the idea behind federated learning. To keep each model on each device unique for those conditions, while learning from the general characteristics of the other models. All of this, while keeping the data private on the device.
At Ekkono our goal is to always provide the best, most accurate, and most sustainable solution in the IoT market. Our research efforts in Federated Learning encompass building lightweight algorithms that can run on any platform, from industrial machinery to microcontrollers; independent of a robust connection between the device and the cloud; and general and personalized for every device under different conditions.
