Google is testing out a new AI solution: Federated Learning Google is testing out a new AI solution: Federated Learning

Google's quest to improve Android devices strongly relies on machine-learning systems, but it seems that the way the company was going about it could use some improvement. This is why, the IT giant has just started testing out a new method called Federated Learning which will not only make machine learning better, but also enhance the users' privacy. At the moment, this new modus operandi is being tested on selected devices with the help of the Gboard app.

In order to explain Federated Learning, you first need to know the current way of doing things: at the moment, Android devices gather relevant usage data on case-by-case basis and send them to the cloud. This new method will actually turn how things are done and place the machine learning system on the device itself instead of the uploading information from the device to the machine learning system.

The machine learning systems will adapt itself to each device and send any modifications back to Google. This is better because the company no longer gets data about the user, just about the modifications to the machine learning system. Moreover, the changes will be made immediately instead of having to wait for an update from the IT giant. Lastly, in order to avoid latency issues, Google will use federated averaging algorithms that reduce the size of the uploaded data and put less strain on the battery. If successful, the Federated Learning method could be another one of Google's tricks that change the way we live.

Referenced Android applications