Computer Vision News - February 2018

“ Find the right balance between professional and family life… ” 18 Computer Vision News Application: UnifyID Vinay Prabhu is UnifyID ’s Principal Machine Learning Scientist and he explains that they harness statistical signal processing and machine learning to solve the issue of a person’s identity. Their technology questions the institution of passwords used for authentication. Instead, they believe in the institution of implicit authentication in which simply one’s entire being acts as a password. UnifyID takes into account factors such as the way a person walks, the constellation of Bluetooth devices in use, and other biometrics. They do this in such a way that the whole experience is pretty seamless, and more importantly, passive. The technology d es not require any additive action on the part of the user. As the person walks, the technology uses a machine learning pipeline that harnesses state-of-the-art deep learning modules to perform deep gait authentication (which they term as GaitNet ). It extracts routine cycles before comparing them with the normative model that the system has learned. The pipeline itself entails harnessing on- device signal processing, spatio- temporal analysis and, of course, deep learning to build a statistical profile of the user’s normative behavior and thus enable passive, seamless authentication. They classify the sensors used by their system to collect this data in two disparate categories. The first category is inertial measurement unit (IMU) sensors , also known colloquially as motion sensors that typically include the triad of tri-axial accelerometers, gyroscopes and magnetometers on the phone. The system also accesses the non-IMU sensors which includes sensors like GPS and Bluetooth sensors that are available on the phone. In some cases, they also support bleeding edge sensors available on the phone such as hydration s nsors. However, the system primarily focuses on the GPS, Bluetooth, and inertial measurement unit sensors. Of course, this means that the user must give them access to all of this information. However, if a user does not want to share certain information such as GPS, the UnifyID system puts a weight of zero in regards to that sensor. This gives complete control to the user. Based on feedback from their users, many seem hesitant to allow access to so many UnifyID is a service that provides authentication based on a person’s movements. Rather than using passwords to gain access, UnifyID identifies a person with unique factors such as how they walk, sit or type. “Insisting on deep learning models in every phase of our pipeline is not something that has paid rich dividends.” Application