Sensor-Based Authentication on Smartphones via Integrating Auxiliary Information

With the widespread use of smartphones, more and more private information is stored on the phone, and the loss or theft of the phone can lead to data leakage, theft of property, and other problems. Traditional active authentication methods based on passwords, faces, fingerprints, etc. authenticate only once at login, which still has some hidden dangers. Authentication methods based on behavioral biometrics such as walking gait, touch screen, keystroke, etc. can provide implicit and continuous authentication services, and thus have been widely studied recently. Considering the limited computational resources of smart devices, we first introduce the state-space model with linear model complexity, i.e., Mamba, to extract deep feature representations from raw signal sequences in the sensor-based authentication task. Then a series of auxiliary information from different domains are proposed and fused to the deep features to enhance the consistent semantic information and ultimately improve the discriminative ability of the model. Our model A-Mamba, i.e., Auxiliary-Mamba, is experimented on two public datasets. The experimental results show that our model outperforms existing approaches.