Towards the next generation of biometric authentication based on kinesthetic intelligence augmented by explainable AI.
Dong Qin · 2024
Biometric user authentication is at the core of multi-factor authentication, and mouse-based biometric authentication comes at no additional cost for most computer systems. This work seeks to improve the mouse-based biometric authentication in three aspects, these improvements are also applicable to other behavioral biometric authentication scenarios that has similar data characteristics. First, to improve authentication accuracy, we propose a new mouse-based user authentication scheme, called MAUSPAD, which uses a novel progress-adjusted dynamic time warping (PADTW) algorithm, along with a segmentation algorithm, to accurately and meaningfully measure the differences between observed data and reference data. By introducing a new concept, which we call \textit{progress}, into standard DTW, the new PADTW can have better control of the warping and mapping process and hence is more suitable for comparing time-stamped spatial sequences such as mouse cursor movements. Second, to improve authenticator transparency, we introduce a feature attribution method called Double-sided Remove and Reconstruct (DoRaR). It addresses two issues that other feature attribution methods may have. Such as artifacts problem, caused by feeding out-of-distribution masked inputs directly to the classifier that was originally trained on natural data points. And Encoding Prediction in the Explanation (EPITE) problem, which the predictor's decisions rely not on the features, but on the masks that selects those features. Therefore, the explanation result become more reliable. Third, to make the behavioral biometric authenticator more robust against adversarial attack, we take advantages of our DoRaR method and propose an eXplainable AI (XAI) based defense strategy. By introducing a feature selector, trained with adversarial sample generator augmented DoRaR, as a filter in front of the original authenticator. It can filter out features that are more vulnerable to adversarial attacks or less contributive to authentication.