Meta-WF: Meta-Learning-Based Few-Shot Wireless Impersonation Detection for Wi-Fi Networks

As the common wireless open medium of the Internet of Things (IoT), Wi-Fi networks are often suffering from impersonation attacks. For the purpose of Wi-Fi impersonation attack detection (WIAD), recently, deep learning-based techniques provide superior detection performance. However, considering the dynamics of the WIAD task, these methods require retraining deep learning network (DNN) through large-scale samples and long training time to adapt to new WIAD tasks. In this letter, we propose a meta-learning-based Wi-Fi impersonation detection (
Meta-WF
) algorithm, which learns knowledge from historical multiple WIAD scenarios and quickly adapts to new WIAD scenarios with few-shot samples. Numerical results show that the proposed algorithm is capable of achieving 98% accuracy for new WIAD tasks by 1-step fine-tune using less than 0.1% (50-shot) of the training set samples.