Enhancing Robust Image Classification with Graph Regularized Adversarial Training

The proposed GReAT method integrates graph-based regularization into the adversarial training process to enhance the robustness of deep learning models against adversarial attacks.