arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2103.08820cs.LGcs.AIcs.CR
收藏

EX-RAY:通过检查微分特征对称性区分神经网络中注入的后门和自然特征

EX-RAY: Distinguishing Injected Backdoor from Natural Features in Neural Networks by Examining Differential Feature Symmetry

Purdue University(普渡大学);Rutgers University(罗格斯大学)

Yingqi Liu, Guangyu Shen, Guanhong Tao, Zhenting Wang, Shiqing Ma, Xiangyu Zhang

摘要:

后门攻击向模型注入恶意行为,使得嵌入有触发器的输入被错误分类为攻击者所需的目标标签。然而,自然特征可能表现得像触发器,一旦嵌入就会导致错误分类。虽然它们是不可避免的,但错误地将它们识别为注入的触发器会导致后门扫描中的错误警告。因此,一个突出的挑战是区分自然特征和注入的后门。我们开发了一种新的对称特征差分方法,该方法可以识别分离两个类的最小特征集。如果相应的触发器包含的特征不同于区分受害者和目标类的特征集,则认为后门被注入。我们在数以千计的模型上评估了这项技术,包括来自TrojAI第2-4轮比赛的干净和木马模型以及ImageNet上的一些模型。现有的后门扫描技术可能产生数百个假阳性(即,被识别为特洛伊木马的干净模型)。我们的技术消除了78-100%的误报(通过最先进的扫描仪ABS),误报率小幅增加0- 30%,实现了17-41%的整体准确性提高,并有助于在排行榜上实现最佳性能。它还提高了其他扫描仪的性能。它优于使用L2距离和归因技术的假阳性去除方法。我们还展示了它在检测许多语义后门攻击方面的潜力。

英文摘要:

Backdoor attack injects malicious behavior to models such that inputs embedded with triggers are misclassified to a target label desired by the attacker. However, natural features may behave like triggers, causing misclassification once embedded. While they are inevitable, mis-recognizing them as injected triggers causes false warnings in backdoor scanning. A prominent challenge is hence to distinguish natural features and injected backdoors. We develop a novel symmetric feature differencing method that identifies a smallest set of features separating two classes. A backdoor is considered injected if the corresponding trigger consists of features different from the set of features distinguishing the victim and target classes. We evaluate the technique on thousands of models, including both clean and trojaned models, from the TrojAI rounds 2-4 competitions and a number of models on ImageNet. Existing backdoor scanning techniques may produce hundreds of false positives (i.e., clean models recognized as trojaned). Our technique removes 78-100% of the false positives (by a state-of-the-art scanner ABS) with a small increase of false negatives by 0-30%, achieving 17-41% overall accuracy improvement, and facilitates achieving top performance on the leaderboard. It also boosts performance of other scanners. It outperforms false positive removal methods using L2 distance and attribution techniques. We also demonstrate its potential in detecting a number of semantic backdoor attacks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑