A NAS-Based TinyML for Secure Authentication Detection on SAGVN-Enabled Consumer Edge Devices

Traditional authentication methods utilize a uniform model to verify all users, which has been proven to be ineffective for the authentication on consumer edge devices of space-air-ground integrated vehicular networks (SAGVN). On the other hand, consumer edge devices in SAGVN have limited resources to load large models for authentication. In order to handle these issues, this paper proposes a Neural Architecture Search-Tiny Machine Learning (NAS-TinyML) method for authentication on SAGVN-enabled consumer edge devices. The NAS-TinyML integrates a feature extraction module derived from NAS and a classification module constructed using Spiking Recurrent Neural Networks (SRNN). It can combine the flexible architecture search capability of NAS, the robustness of SRNN in handling time-series data, and a macro-level search mechanism to enhance model search efficiency. The validation on various scenario datasets shows that NAS-TinyML is superior to several state-of-the-art methods, including GeneticNAS, Drnas, NASP, NetAdapt, SMASH, and SNAS. Furthermore, the deployment of pre-trained NAS-TinyML on different consumer edge devices demonstrates its effectiveness.