GANDACOG: Implicit Mobile User Authentication in Multi Environments With Scarce Data

With the widespread adoption of mobile devices, user authentication has become a critical component in ensuring personal information security. Mobile devices (such as smartphones) are indispensable in IoT scenarios, and the security authentication of mobile device users has become an increasing focus. Motion sensor-based authentication methods can address the privacy leakage and high costs associated with traditional authentication methods (such as face recognition, fingerprints, and passwords) by implicitly collecting user interaction data. However, existing motion sensor-based user authentication still faces challenges, such as limited training samples, poor adaptability to diverse scenarios, and high computational overhead, making it difficult to meet the demands for efficiency, low power consumption, and high security in real-world applications. To address these challenges, we propose a novel mobile user authentication framework, GANDACOG, which integrates multiple optimization strategies. Specifically: 1) we design AUTHGANS, a generative model that produces high-quality synthetic data to alleviate the issue of insufficient training samples; 2) we introduce a multihead differential attention mechanism to effectively suppress noise interference, improve adaptability to different usage scenarios, and optimize model structure during training to reduce computational overhead; and 3) we incorporate AuthFusion for model distillation, which significantly reduces on-device computational requirements while ensuring high authentication accuracy. Experiments carried out on a real-world dataset that contains 1513 users demonstrate that, compared to state-of-the-art authentication methods, the accuracy of GANDACOG can be improved by more than 10% in different scenarios, and the compression efficiency can exceed 70%.