Three years Google has implemented projects and as innovative as aggressive strategies to promote new techniques of machine learning, and now, the company wants to use the smartphone for each user on the Android platform to improve their algorithms of artificial intelligence.
A publication on the official blog of Google research has revealed all the details on this new initiative, called “Learning land” (Federated Learning), with which the company seeks to decentralize training and improvement of artificial intelligence processes so that all happen within the own smarpthone.
Now, Google rather than collect data from users in one place (your own servers) and apply its algorithms of training there, to then return to the mobile device improvements, implement the teaching process will occur directly in each user smarpthone.
This new scheme would become much more agile, private and personalized learning and adaptation of its artificial intelligence process. To prove it, in this first stage of release of the model of learning, Google will use your application’s keyboard Gboard, as a way to study the functioning of this dynamic.
Now, when Gboard shows users search suggestions based on your prior record of messages, the application will remember suggestions were used and which ignored. This information will be used to customize the application algorithms directly on users ‘ phones.
To achieve this trick Gboard has now incorporated a light version of TensorFlow, the automatic learning of Google software.
If the experiment works, we can expect a deeper invasion of this model in our smartphones.
Follow us and comment on Facebook.