Jieyu Lin, Wei Liu, Jiajie Zou, Nai Ding: Training NLI Models Through Universal Adversarial Attack. CCL 2023: 306-324