Interpreting Adversarially Trained Convolutional Neural Networks
This work provides insights into the interpretability of adversarial training for CNNs, which is incremental but useful for researchers in machine learning security and robustness.
The study investigated how adversarially trained convolutional neural networks (AT-CNNs) recognize objects, finding that adversarial training reduces texture bias and promotes shape-biased representations, leading to improved robustness compared to normally trained CNNs.
We attempt to interpret how adversarially trained convolutional neural networks (AT-CNNs) recognize objects. We design systematic approaches to interpret AT-CNNs in both qualitative and quantitative ways and compare them with normally trained models. Surprisingly, we find that adversarial training alleviates the texture bias of standard CNNs when trained on object recognition tasks, and helps CNNs learn a more shape-biased representation. We validate our hypothesis from two aspects. First, we compare the salience maps of AT-CNNs and standard CNNs on clean images and images under different transformations. The comparison could visually show that the prediction of the two types of CNNs is sensitive to dramatically different types of features. Second, to achieve quantitative verification, we construct additional test datasets that destroy either textures or shapes, such as style-transferred version of clean data, saturated images and patch-shuffled ones, and then evaluate the classification accuracy of AT-CNNs and normal CNNs on these datasets. Our findings shed some light on why AT-CNNs are more robust than those normally trained ones and contribute to a better understanding of adversarial training over CNNs from an interpretation perspective.