Tech breakdown: What is Federated Learning?
Google has invented an approach to learn about users without their data ever leaving their device. Federated Learning is a form of machine learning that enables mobile phones to collaboratively learn a shared prediction model while keeping all the training data on device. This means that while it won’t collect data from its users, the technology will make products smarter over time.
Let’s break down how this AI tech works. When you use Google’s keyboard, Gboard, Federated Learning recognizes the words you use and stores this information. This is how your keyboard learns new words you use that aren’t in the dictionary yet.
First, your device downloads a generic learning model. As you use your phone, federated learning personalizes and improves the current model then computes a summary of the changes. It then updates based on things you and other people have been typing, and uploads the updated model back to the server. You might think that this will take a toll on your smartphone’s battery, but it won’t. Federated Learning will occur only when your phone is charging, on Wi-Fi, and idle. This will provide a global improvement to the model, making it work better for everybody.
Again, this learning innovation does not collect your data; it only trains results such as the words you type. All the training data remains on your device, and no individual updates are stored in the cloud.
This innovation allows for smarter models, lower latency, and less power consumption, all while ensuring privacy.
Google has yet to announce which next apps or services will be tested with Federated Learning.However, Google AI lead Blaise Agüera y Arcas shared that the tech giant is working on expanding it to the ecosystem.
For more information about Federated Learning, visit federated.withgoogle.com.