SBIR Phase I: IMPULSE: Implicit private authentication with ultrasonic signals from mobile ecosystem

Unknot.Id Inc., Orlando FL

Investigators

Abstract

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project will result from enabling a secure, scalable, adversarial-resistant, fully private authentication for mobile ecosystems using behavioral biometrics, thereby enabling consumer control of digital privacy. An immediate impact will be the mitigation of mobile based fraud as well as financial burden for companies struggling to build GDPR compliant one-to-many biometric technologies. Besides fraud detection applications, the technology developed to infer multiple user biosignatures passively from smartphones will enable the development of non-invasive and cost-effective telehealth solutions, thereby advancing the state-of-the-art monitoring of health using wearable physiological monitors, Wi-Fi, and radar sensors. Impacts of the project more broadly will result from various education and outreach activities, including internships, mentoring, and participation in seminars and conferences that encourage diversity, and by establishing research and education collaboration with faculty and students in underrepresented communities. This Small Business Innovation Research Phase I project addresses the long-standing challenge in securing one-to-many biometric systems, widely used in hospitals, border control infrastructure and more. Recent cyber security attacks and privacy violations makes it critical to develop an adversarial resistant technology that enables secure biometric authentication at scale, while at the same time providing rigorous privacy assurance to users. Consequently, this effort will advance the state-of-the-art in key fronts: (1) a new class of sufficiently distinguishable unique behavioral biosignatures extracted passively and non-invasively from high frequency sound waves using deep learning models; (2) efficient privacy-preserving algorithms to enable fully private biometrics while ensuring data utility and faster similarity search; (3) novel techniques to stress-test the machine learning framework in order to discover new attack vectors in behavioral biometrics. This technology will enhance biometrics with user behavioral biosignatures which are exclusively and passively derived from audio sensors on smartphones, using deep transfer learning algorithms, to thwart spoofing and adversarial attacks; furthermore, it will transform sensitive biometric identifiers into revocable and anonymized identifiers, with local differential privacy and hashing-based privacy-preserving algorithms. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.