SemRoDe: Macro Adversarial Training for Robust Language Models

Adversarial training with distance alignment enhances robustness in language models.