Federated learning is a privacy-enhancing technology that we use to improve models on-device without sending your raw data to Google servers. Google uses federated learning to improve device activation such as “Hey Google” on Google Assistant-enabled devices.
When you ask, "Hey Google, what’s the weather tomorrow?" an on-device model detects that you said, “Hey Google,” and then sends your query to Google Assistant.
Also, on supported smart home devices with cameras, on-device models can store and use images to detect activation through gestures, gaze, and lip movement.
Your devices might activate unintentionally or trigger when you didn't say, “Hey Google.” In other cases, it might not trigger when you did say, “Hey Google.”
We now use federated learning to refine Assistant's activation technologies to reduce misactivations and misses.
Learn more about federated learning from Google AI’s online comic.
Manage settings & your data
By default, the use of federated learning to improve Assistant's activation technologies is turned off.