Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study addresses the challenges of applying keystroke dynamics for user authentication across different devices by employing inductive transfer learning to adapt typing patterns from one device to another. By combining adapted keystroke data with limited training data from the secondary device, the researchers trained a robust binary classifier that effectively mitigates distribution drift issues. The proposed system achieved an equal error rate of 14.2% in cross-device scenarios, outperforming existing state-of-the-art methods.
Transfer learning can bridge the gap in keystroke dynamics, achieving a 14.2% error rate in cross-device authentication—better than any previous approach.
Keystroke dynamics (typing patterns) can be used as a behavioural biometric modality for user authentication, with applications such as fraud prevention. While the modality has been shown to work well for single device authentication, its application to cross-device scenarios is more challenging. Dynamics learned on one device (eg., phone) may not be directly applicable to authentication on a secondary device with a different form factor (eg., tablet) due to changes in typing patterns that can lead to distribution drifts. To address this, we propose a cross-device user authentication system based on inductive transfer learning, where keystroke dynamics learned on one device are adapted to a secondary device. The adapted data is then combined with necessarily limited training data for the secondary device, which is used to robustly train a binary classifier. Furthermore, an extended set of keystroke features is used to better capture discriminative dynamics. Experiments on the BBMAS dataset show that proposed system achieves an equal error rate of 14.2% for the cross-device scenario, surpassing state-of-the-art methods.