All available observations are fused together to generate a data set for each user that characterizes their behavior. Background motion and location as well as phone interaction characteristics make up the core input streams.
Deep neural networks are trained on these data streams to find the key aspects that distinguish this user from all others. Where someone carries a device, the length of their stride, the difference in behavior from home to work, all of these parameters are taken into account.
The same fusion and learning methods are also used to identify key events that coincide with fraudulent behavior. Understanding and detecting when devices change hands gives special insight into fraudulent activity.
Once you use two-factor authentication, you go through more hoops, you're more aware of security, you're spending a certain amount of time each month [on it]. […] In the future we need to make the secure systems easier to use. Getting a nice user interface to a secure system is the art of the century.
Inventor of the World Wide Web