In 2010,Lee et al proposed two simple and efficient three-party password-authenticated key exchange protocols that had been proven secure in the random oracle model.They argued that the two protocols could resist offl...In 2010,Lee et al proposed two simple and efficient three-party password-authenticated key exchange protocols that had been proven secure in the random oracle model.They argued that the two protocols could resist offline dictionary attacks.Indeed,the provable approach did not provide protection against off-line dictionary attacks.This paper shows that the two protocols are vulnerable to off-line dictionary attacks in the presence of an inside attacker because of an authentication flaw.This study conducts a detailed analysis on the flaw in the protocols and also shows how to eliminate the security flaw.展开更多
基金Supported by the Natural Science Foundation of Jiangsu Province (Key Program) (BK2011023)
文摘In 2010,Lee et al proposed two simple and efficient three-party password-authenticated key exchange protocols that had been proven secure in the random oracle model.They argued that the two protocols could resist offline dictionary attacks.Indeed,the provable approach did not provide protection against off-line dictionary attacks.This paper shows that the two protocols are vulnerable to off-line dictionary attacks in the presence of an inside attacker because of an authentication flaw.This study conducts a detailed analysis on the flaw in the protocols and also shows how to eliminate the security flaw.
文摘对抗样本通过对输入进行少量扰动使得神经网络产生误判,是衡量深度学习模型鲁棒性的重要手段。中文对抗样本存在覆盖情况不广、质量不高、攻击效果不强等问题。为解决以上问题,论文提出一种字符级文本对抗样本生成方法SCGA(Similar Character Generation Adversarial Example)。该方法对汉字的读音和形状进行相似度计算并构建音形相近字典,然后定位到关键字,利用替换攻击生成对抗样本。最后在黑盒下进行攻击实验。在多个数据集上与多个方法对比验证了方法的有效性。该方法所生成的对抗样本质量较高且能有效的误导模型的分类结果。
