Event Description
Learning of Identity from Behavioral Biometrics for Active Authentication on Desktop Computers and Mobile Devices
Ph.D. Dissertation Defense
Presenter: Alex Fridman
Advisors: Drs. Steven Weber and Moshe Kam
Abstract: In this work, we investigate the problem of active authentication on desktop computers and mobile devices. Active authentication is the process of continuously verifying a person's identity based on the cognitive, behavioral, and physical aspects of their interaction with the device. We consider several representative modalities including keystroke dynamics, mouse movement, application usage patterns, web browsing behavior, GPS location, and stylometry. We implement a binary classifier for each modality and organize the classifiers as a parallel binary decision fusion architecture. The decisions of each classifier are fed into a decision fusion center (DFC) which applies the Chair-Varshney fusion rule to generate a global decision. The DFC minimizes the probability of error using estimates of each local classifier's false rejection rate (FAR) and false acceptance rate (FRR). We test our approach on two large datasets of 67 desktop computer users and 200 mobile device users. We are able to characterize the performance of the system with respect to intruder detection time and to quantify the contribution of each modality to the overall performance. |
