MGRAuth: Sensor-Based Continuous Authentication With a Mixture-of-Experts Gated-Relation Autoencoder
IEEE Internet of Things Journal · 15.04.2026 · 13 (8) · 17482-17496
DOI: 10.1109/jiot.2026.3662267 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Internet of Things Journal, erschienen am 15.04.2026. Verfasst von Yantao Li und 4 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 55 Literaturangaben. Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
With the increasing reliance on mobile devices for storing sensitive personal information, continuous authentication has become essential for providing persistent security beyond one-time login. Sensor-based continuous authentication leverages behavioral biometrics collected from built-in sensors; however, existing approaches often suffer from limited robustness due to multisensor heterogeneity, diverse behavioral patterns, and performance degradation under noisy and dynamic usage conditions. To address these challenges, we propose MGRAuth, a novel sensor-based continuous Authentication system that integrates a Mixture-of-experts (MoE) architecture and a Gated-Relation network (GRN) within an Autoencoder model. MGRAuth incorporates a sensor-level MoE (SL-MoE) to model modality-specific characteristics from heterogeneous sensor inputs and a pattern-level MoE (PL-MoE) to capture diverse behavioral tasks and their temporal transitions. In addition, the GRN enhances identity-related rhythmic features while mitigating noise-induced disturbances through gated relational modeling. Authentication is performed by reconstructing legitimate users’ sensor data and comparing the reconstruction error (RE) against a predefined threshold. Experimental results on our dataset demonstrate that MGRAuth achieves an accuracy of 99.34% and an equal error rate (EER) of 0.64% with 87 unseen users, outperforming representative state-of-the-art methods. These results indicate that MGRAuth provides a robust and high-accuracy solution for sensor-based continuous authentication on mobile devices.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Li, Y., He, Z., Zhao, W., Huang, H., & Deng, S. (2026). MGRAuth: Sensor-Based Continuous Authentication With a Mixture-of-Experts Gated-Relation Autoencoder. IEEE Internet of Things Journal, 13(8), 17482-17496. https://doi.org/10.1109/jiot.2026.3662267
Vancouver
Li Y, He Z, Zhao W, Huang H, Deng S. MGRAuth: Sensor-Based Continuous Authentication With a Mixture-of-Experts Gated-Relation Autoencoder. IEEE Internet of Things Journal. 2026;13(8):17482-17496. doi:10.1109/jiot.2026.3662267
BibTeX
@article{Li2026,
author = {Li, Yantao and He, Zhenglu and Zhao, Wenyan and Huang, Hongyu and Deng, Shaojiang},
title = {MGRAuth: Sensor-Based Continuous Authentication With a Mixture-of-Experts Gated-Relation Autoencoder},
journal = {IEEE Internet of Things Journal},
year = {2026},
volume = {13},
number = {8},
pages = {17482--17496},
doi = {10.1109/jiot.2026.3662267}
}
EUVIMED