On 6th April 2017, Google discussed their new research on Gboard which is the virtual keyboard application for Android. This research is all about training the artificial neural networks on data which are located locally on the mobile devices.
“When Gboard shows a suggested query, your phone locally stores information about the current context and whether you clicked the suggestion. Federated Learning processes that history on-device to suggest improvements to the next iteration of Gboard’s query suggestion model,” said Google research scientists Brendan McMahan and Daniel Ramage in a blog post.
They expect to use the similar technology for improving the photo ranking which is basically based on what is the type of photos people look for, share or delete.In the past, Apple had announced that they were to deploy the concept called as differential privacy in order to turn out Quicktype and Emoji suggestion in iOS10.
Firstly Google makes a suggestion and from the existing data, then it summarizes the changes taking place which is called as an update. All the update is in an encrypted format which Google’s averages with other updates. Google can only decrypt its average update and the actual data will always stay local.
In the past, Google has already explored about the randomized aggregatable privacy-preserving ordinal response. Google is now going more deeper with the prominent app Gboard which currently has around 500 million to 1 billion app installation and is available in Google Play Store.