类条件神经偏振器:通过净化被污染特征实现轻量且有效的后门防御
Class-Conditional Neural Polarizer: A Lightweight and Effective Backdoor Defense by Purifying Poisoned Features
AI总结:
提出类条件神经偏振器(CNPD)后门防御方法,通过融合模型预测标签与特征净化中毒样本,无需标签估计,兼顾防御效果与良性性能,提供三种实现变体。
摘要:
近期研究凸显了深度神经网络对后门攻击的脆弱性,在这种攻击中,模型被操纵以依赖中毒样本中嵌入的触发器,尽管样本同时包含良性信息和触发器信息。虽然已有多种防御方法被提出,但它们往往难以在缓解后门攻击与保持良性性能之间取得平衡。在这项工作中,受光学偏振器概念的启发——该器件允许特定偏振的光波通过而滤除其他偏振——我们提出了一种轻量级的后门防御方法NPD。该方法将神经偏振器(NP)作为中间层集成到被攻陷的模型中,实现为通过双层优化优化的轻量级线性变换。可学习的NP从中毒样本中滤除触发器信息,同时保留良性内容。尽管有效,但通过实证研究我们发现,当用于净化的目标标签估计不准确时,NPD的性能会下降。为了解决这一局限,同时利用定向对抗缓解的潜力,我们提出了基于类条件神经偏振器的防御方法(CNPD)。其关键创新在于一个融合模块,该模块将被后门攻击的模型预测的标签与待净化的特征相整合。这一架构本质上模仿了定向对抗防御机制,而无需使用NPD中所需的标签估计。我们提出了CNPD的三种实现:第一种是r-CNPD,它为每个类别训练一个复制的NP层,并在推理时根据被后门攻击的模型预测的类别选择适当的NP层进行防御。为了高效处理大量类别,设计了两种变体:e-CNPD,将类别信息作为附加特征嵌入;以及a-CNPD,利用类别信息引导网络注意力。
英文摘要:
Recent studies have highlighted the vulnerability of deep neural networks to backdoor attacks, where models are manipulated to rely on embedded triggers within poisoned samples, despite the presence of both benign and trigger information. While several defense methods have been proposed, they often struggle to balance backdoor mitigation with maintaining benign performance.In this work, inspired by the concept of optical polarizer-which allows light waves of specific polarizations to pass while filtering others-we propose a lightweight backdoor defense approach, NPD. This method integrates a neural polarizer (NP) as an intermediate layer within the compromised model, implemented as a lightweight linear transformation optimized via bi-level optimization. The learnable NP filters trigger information from poisoned samples while preserving benign content. Despite its effectiveness, we identify through empirical studies that NPD's performance degrades when the target labels (required for purification) are inaccurately estimated. To address this limitation while harnessing the potential of targeted adversarial mitigation, we propose class-conditional neural polarizer-based defense (CNPD). The key innovation is a fusion module that integrates the backdoored model's predicted label with the features to be purified. This architecture inherently mimics targeted adversarial defense mechanisms without requiring label estimation used in NPD. We propose three implementations of CNPD: the first is r-CNPD, which trains a replicated NP layer for each class and, during inference, selects the appropriate NP layer for defense based on the predicted class from the backdoored model. To efficiently handle a large number of classes, two variants are designed: e-CNPD, which embeds class information as additional features, and a-CNPD, which directs network attention using class information.
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。