Interpreting Adversarial Examples with Attributes
This addresses the need for interpretability in adversarial robustness for computer vision, but it is incremental as it builds on existing attribute-based methods.
The paper tackles the problem of interpreting adversarial examples in deep vision systems by using attributes to justify network decisions for both clean and adversarial inputs, showing results through experiments on three benchmark datasets.
Deep computer vision systems being vulnerable to imperceptible and carefully crafted noise have raised questions regarding the robustness of their decisions. We take a step back and approach this problem from an orthogonal direction. We propose to enable black-box neural networks to justify their reasoning both for clean and for adversarial examples by leveraging attributes, i.e. visually discriminative properties of objects. We rank attributes based on their class relevance, i.e. how the classification decision changes when the input is visually slightly perturbed, as well as image relevance, i.e. how well the attributes can be localized on both clean and perturbed images. We present comprehensive experiments for attribute prediction, adversarial example generation, adversarially robust learning, and their qualitative and quantitative analysis using predicted attributes on three benchmark datasets.