A Biometric Testbed to Support NCATÕS Biometric Research and Education

This proposal, involving the Computer Science (CS) and Computational Science and Engineering (CSE) Departments at North Carolina A&T State University (NCAT), requests equipment for a biometric testbed for research and education supported by the ARO and others. Biometrics are the characteristics of humans used to recognize them. Biometric-based authentication is popular since the modalities used in it are difficult to replicate and are unique to individuals. Physical biometrics (such as facial images) are based on intrinsic physical traits while behavioral biometrics (e.g., swipe patterns) are based on behavioral traits. Standard, one-time authentication is done only at the start of a session while active authentication (typically with behavior biometrics) verifies that the user in control is the originally authenticated user. The current ARO grant, ÒTrustworthy, Privacy Enhanced and Secure Cyber Identity Framework,Ó 09/01/2015 to 08/31/2019, aims to enhance capabilities in cyber-identity research at the DoD and NCAT. Four thrusts were proposed. (1) Develop a theoretical framework for identity, leading into a computational framework. For disclosing identity, we use the WebID decentralized authentication protocol. (2) Develop a multi-factor active authentication system. (3) Mitigate biometric-based replay attacks, where an attacker captures packets and replays them later to gain unauthorized access. (4) Secure cloud data using biometric-based authentication, with aspects taken from (2) and (3). Biometrics are key, but processing biometric data is computationally intensive, even infeasible, without the required equipment. The following thrusts extend the critical parts of the grant but require the requested equipment. Deep learning is used extensively; a major concern is training these in reasonable time, for which GPUs are critical. I. Authentication using deep learning, including training convolutional neural nets (CNNs). II. Active authentication using touch gestures, two kinds of artificial immune systems (AISs). III. Mitigating biometric-based presentation (spoofing) attacks. We aim to detect presentation attacks using CNNs. IV. Human activity recognition. We use deep learning with biometric motion data from mobile devices to identify what a person is doing and whether it is in fact the authorized person. V. Incorporating biometrics and policies into the WebID protocol. Regarding education, we have identified four relevant core areas in both the CS and the CSE Departments. We propose two new graduate courses cross-listed in the two departments: an introduction to biometrics and a course on deep learning. We also plan to have a seminar series and a special interest group in deep learning. The requested equipment will complement the traditional equipment in our biometrics research lab, providing a biometrics testbed. All equipment will have a life of five or more years. We present the requests under five categories and identify the thrusts and the approximate cost for each category; the overall request is about $199K. ¥ Deep learning workstations and monitors ($88K, 44% of the request). Thrusts I-IV use deep learning, which needs GPU-based processing ¥ Regular Workstations and Monitors ($44K). Students use these workstations for general biometric-related research and course work. ¥ Servers ($27K). For Thrust V, continuous cloud authentication (Thrust II), Thrust III setup ¥ Situational Awareness and Display Devices ($6K). For Thrust IV ¥ Biometric Sensors ($35K). All thrusts, and to capture images & build our own datasets