A Dynamic Heuristic Search for Textual Adversarial Attacks

Adversarial attacks represent a significant threat to the robustness of deep learning models by exploiting their vulnerabilities. Recent advancements have given rise to high quality attack techniques, some of which are strategically employed to enhance model robustness via adversarial training and defense strategies. This paper introduces a novel heuristic search algorithm designed for the generation of adversarial samples through iterative synonym ranking. Our approach distinguishes itself through a superior success rate, lower accuracy under attack, surpassing many existing state-of-the-art methods, while maintaining a acceptable query count. Through extensive experiments, we demonstrate the efficacy of our approach in enhancing deep learning models against adversarial attacks through adversarial training.