Logit 间隔排斥用于后门防御
Logit-Margin Repulsion for Backdoor Defense
针对量化/剪枝后才激活的条件后门，LMR 用少量干净样本配合 logit-margin 约束放大后门类与良性类的间隔，再按通道选择性剪枝去除后门通路；做模型压缩部署或需要清洗第三方预训练权重时，这套「间隔约束+通道定位」流程可直接复用。
原文依据 LMR uses a small set of clean samples and combines selective cross-entropy with a logit-margin constraint to enlarge the gap between the backdoor class and benign classes.
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论后门攻击对深度神经网络构成重大威胁。近期研究表明，攻击者可利用量化、剪枝等模型压缩手段植入条件后门。此类后门在原始模型中处于休眠状态，只有在模型经过特定操作后才会被激活，因而具有极强的隐蔽性且难以检测。传统防御方法难以应对这类攻击，而专门针对条件后门设计的防御方法也难以处理传统后门攻击。为应对这些挑战，我们提出一种通用防御方法，称为 Logit 间隔排斥（Logit Margin Repulsion，LMR）。LMR 使用少量干净样本，将选择性交叉熵与 logit 间隔约束相结合，以扩大后门类别与良性类别之间的间隔；随后通过选择性剪枝移除与后门行为相关的通道，从而实现强力的后门净化。在多种 CNN 与视觉 Transformer 上的大量实验表明，即使干净数据量极其有限（0.1%），LMR 仍能有效缓解传统后门攻击与条件后门攻击。实现代码已公开于 https://github.com/Trusted-LLM/LMR。
英文原文摘要 1293 字符
Backdoor attacks pose a significant threat to deep neural networks. Recent studies have shown that model compression, such as quantization and pruning, can be exploited by attackers to implant conditional backdoors. Such backdoors remain dormant in the original model but are activated after the model undergoes specific operations, making them highly stealthy and difficult to detect. Traditional defense methods struggle to counter this type of attack, while defenses specifically designed for conditional backdoors also have difficulty handling traditional backdoor attacks. To address these challenges, we propose a universal defense method, termed Logit Margin Repulsion (LMR). LMR uses a small set of clean samples and combines selective cross-entropy with a logit-margin constraint to enlarge the gap between the backdoor class and benign classes. It then removes channels associated with backdoor behavior through selective pruning, thereby achieving strong backdoor purification. Extensive experiments on a variety of CNNs and Vision Transformers demonstrate that, even with an extremely limited amount of clean data (0.1%), LMR can effectively mitigate both traditional and conditional backdoor attacks. The implementation is publicly available on https://github.com/Trusted-LLM/LMR.