一句话：提出用蜕变测试评估 DNN 解释器（解释方法）的鲁棒性，以应对同一分类案例下解释结果不一致的问题。
深度神经网络（DNN）的黑盒特性使其决策难以理解，限制了其在关键任务中的应用。近期许多解释器（解释方法）通过显著图（saliency map）提供相关特征，以提升 DNN 的透明度。然而，不同解释器对同一分类案例可能给出不同的解释结果，这促使作者对解释器开展鲁棒性评估。论文更进一步，采用蜕变测试（metamorphic testing）来评估解释器，利用蜕变关系考察解释结果在输入变换下是否保持一致或符合预期。摘要未给出具体实验结论，核心贡献在于将蜕变测试引入 DNN 解释器的鲁棒性评估。
The black-box nature of the Deep Neural Network (DNN) makes it difficult for people to understand why it makes a specific decision, which restricts its applications in critical tasks. Recently, many interpreters (interpretation methods) are proposed to improve the transparency of DNNs by providing relevant features in the form of a saliency map. However, different interpreters might provide different interpretation results for the same classification case, which motivates us to conduct the robustness evaluation of interpreters.
DOI 原文 ·
@inproceedings{paperbot7402,
title = {One step further: evaluating interpreters using metamorphic testing},
author = {Ming Fan and Jiali Wei and Wuxia Jin and Zhou Xu and Wenying Wei and Ting Liu},
booktitle = {ISSTA},
year = {2022},
doi = {10.1145/3533767.3534225}
}