Google, the company that rules the internet, is now testing
a method to leverage its machine learning with its ubiquitous presence on
mobile devices. The new model they're employing is called Federated Learning,
and it hopes to apply artificial intelligence to Google's services on Android
without compromising user privacy.
The Federated Learning model would be downloaded directly
onto a mobile device. It would then learn from data gathered by Google's apps
and services, and apply this knowledge as an update to improve the experience
on your phone.
Fairly straightforward so far, right? But the trick would be
gathering and using this data without compromising the user's privacy. To
accomplish that, the information is encrypted, anonymized, and aggregated with
other users' data before being analyzed in the cloud. The combined user data is
then used to improve the base model of Google's ecosystem.
But before the aggregated data is used to improve Google
services on the whole, a bit of machine learning is applied locally to enhance
your personal experience. This act takes place strictly on your phone, and is
not shared with Google's cloud-based AI until after it's been anonymized and
combined with other users' data.
The improvements made by the local AI will happen
immediately, "powering experiences personalized by the way you use your
phone."
On that front, the research currently being done involves
the Gboard virtual keyboard app on Android. Following the Federated learning
model previously described, Venture Beat explains:
To
sum it up, research scientists Brendan McMahan and Daniel Ramage stated in a blog post that "Federated Learning allows for smarter
models, lower latency, and less power consumption, all while ensuring privacy."
Not edited by
us
