I’m currently developing a flower application that runs the server on a smart home gateway (RPi 5 DiyHue) and is trained on different IoT devices (Ip camera, smart home control panel, etc.). I want to know if it’s able to use the clients for prediction as well or if it’s better to send the input data from the client to the server and only predict there.
hi @neonduck,
Yes, clients can do prediction, and that is usually the better choice.
If an IoT device can run inference at acceptable latency, keep prediction on the client. This gives better privacy, lower latency, and less network traffic. The Flower server does not need to be involved in inference and can focus on coordinating training and model updates.
1 Like
This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.