Optimization of Meta-Learning Algorithms Based on Adversarial Training in Vehicular and Mobile Communications

This research aims to enhance the robustness of machine learning models while improving generalization performance by incorporating meta-learning algorithms into adversarial training methods. Past work has typically found that adversarial training (AT) improves model robustness, but often at the expense of generalization. In this study, we propose a new meta free adversarial training (Meta Free AT) method through meta-parameter optimization and negative feedback mechanism, which optimizes the simultaneous constraints on the perturbation and classifier parameters in each iteration, thus reducing the proportion of incorrectly predicted samples in the model, and effectively narrowing the generalization gap between the training and test samples. This approach can be applied to complex wireless communication scenarios, particularly in vehicular communication, mobile communication, and service domains. Through the combination of meta learning and adversarial training, the model is able to maintain robustness and good generalization ability in complex environments, such as spectrum sharing, interference cancellation and channel propagation, which optimizes the command and control of the wireless communication system, and improves the overall performance of the communication system.