不用扰动输入，通过开关模型部分实现更可靠的解释。
Activation-Deactivation: A General Framework for Robust Post-hoc Explainable AI
- 用开关模型层替代输入扰动，避免分布外样本问题。
- 无需重新训练，可直接接入任意训练好的CNN，解释更鲁棒且可迁移。
- 在多架构、多数据集上优于当前最先进方法。
基于扰动的可解释性方法因依赖分布外的变异输入而受到批评，引发对其解释质量的质疑。本文提出一种新型前向传播范式——激活-去活化（Activation-Deactivation, AD），无需对输入进行扰动。AD将输入特征的扰动替换为关闭对应模型部分的机制。我们实现了ConvAD，一种适用于CNN的AD近似算法。ConvAD是即插即用的机制，可轻松集成到任意训练好的CNN中，无需额外训练即可生成更鲁棒、更可迁移的解释。我们在多个架构、数据集、方法和扰动策略下进行了评估，结果表明ConvAD在解释质量上优于现有最先进方法。
原文摘要 · Abstract (English)
Perturbation-based explainability methods face criticism due to their reliance on out-of-distribution mutants. This raises doubts about the quality of the explanations. In this paper, we introduce a novel forward pass paradigm, Activation-Deactivation (AD), which obviates the need for perturbation of the input. AD replaces perturbation of input features with switching off parts of the model corresponding to to the intended perturbations. We implement ConvAD, an AD approximation algorithm for CNNs. ConvAD is a drop-in mechanism that can be easily added to any trained CNN and, without any additional training, generates more robust and more transferable explanations. We provide evaluation results across multiple architectures, datasets, methods and perturbation strategies, demonstrating the superior quality of ConvAD compared to the SOTA.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。