Smartphone Authentication with Lightweight Deep Learning

Over the past decade, machine learning methods have found their way into a large variety of computer security applications, including accurate spam detection, scalable discovery of new malware families, identifying malware download events in vast amounts of web traffic, detecting software exploits, blocking phishing web pages, and preventing fraudulent financial transactions, just to name a few. At the same time, machine learning methods themselves have evolved. In particular, Deep Learning methods have recently demonstrated great improvements over more “traditional” learning approaches on a number of important tasks, including image and audio classification, natural language processing, machine translation, etc. Deep learning has faced an exponential growth over the last decade. Today, it has found use in a wide range of applications, security being one of those. In this project we try to exploit deep learning methodologies to facilitate user authentication on smartphones. Smartphone authentication has explored a variety of domains like iris recognition (latest Samsung Galaxy S8), face recognition (iPhoneX), fingerprints (traditional iPhones), PINs and pattern recognitions (traditional smartphones). In this project, we try to develop a lightweight authentication scheme based on sensor data which continually authenticates the user at regular intervals of time, thus providing greater security.