标题： 隔离触发器：检测和消除逃避自适应后门 显示英文标题

标题： Isolate Trigger: Detecting and Eradicating Evade-Adaptive Backdoors

摘要： 所有当前针对深度学习模型的后门攻击检测都属于非必要特征(NEF)类别，这些检测方法专注于对抗简单且高效的垂直类别后门——触发器小、少且不与源特征重叠。 逃避自适应后门(EAB)攻击已经避开了NEF检测并提高了训练效率。 我们引入了一个精确、高效且通用的检测和防御框架，称为隔离触发器(IsTr)。 IsTr旨在通过打破源特征的障碍来找到隐藏的触发器。 因此，它研究了后门触发的本质，并使用步骤和差分中间切片作为组件来更新过去的距离和梯度理论。 IsTr在模型中也起到积极作用，无论是否存在后门。 例如，准确地发现并修复自动驾驶中由于故意或无意训练导致的错误识别。 在各种任务上的广泛实验，包括MNIST、人脸识别和交通标志识别，证实了IsTr的高效率、通用性和精确性。 我们对IsTr在一系列六种EAB攻击中的有效性进行了严格评估，包括Badnets、Sin-Wave、多触发器、SSBAs、CASSOCK、HCB。 这些对策都没有被避开，即使攻击组合在一起且触发器与源特征重叠。 显示英文摘要

摘要： All current detection of backdoor attacks on deep learning models fall under the category of a non essential features(NEF), which focus on fighting against simple and efficient vertical class backdoor -- trigger is small, few and not overlapping with the source. Evade-adaptive backdoor (EAB) attacks have evaded NEF detection and improved training efficiency. We introduces a precise, efficient and universal detection and defense framework coined as Isolate Trigger (IsTr). IsTr aims to find the hidden trigger by breaking the barrier of the source features. Therefore, it investigates the essence of backdoor triggering, and uses Steps and Differential-Middle-Slice as components to update past theories of distance and gradient. IsTr also plays a positive role in the model, whether the backdoor exists. For example, accurately find and repair the wrong identification caused by deliberate or unintentional training in automatic driving. Extensive experiments on robustness scross various tasks, including MNIST, facial recognition, and traffic sign recognition, confirm the high efficiency, generality and precision of the IsTr. We rigorously evaluated the effectiveness of the IsTr against a series of six EAB attacks, including Badnets, Sin-Wave, Multi-trigger, SSBAs, CASSOCK, HCB. None of these countermeasures evade, even when attacks are combined and the trigger and source overlap.