Abstract: Few-shot learning (FSL) facilitates a variety of computer vision tasks yet remains vulnerable to adversarial attacks. Existing adversarially robust FSL methods rely on either visual ...