Adversarial Examples Are Not Bugs, They Are Features
This addresses a foundational issue in AI safety and robustness for researchers and practitioners, offering a novel perspective rather than an incremental improvement.
The paper tackles the problem of understanding why adversarial examples exist in machine learning, demonstrating that they are caused by non-robust features—predictive but brittle patterns in data—and provides a theoretical framework to explain their pervasiveness in standard datasets.
Adversarial examples have attracted significant attention in machine learning, but the reasons for their existence and pervasiveness remain unclear. We demonstrate that adversarial examples can be directly attributed to the presence of non-robust features: features derived from patterns in the data distribution that are highly predictive, yet brittle and incomprehensible to humans. After capturing these features within a theoretical framework, we establish their widespread existence in standard datasets. Finally, we present a simple setting where we can rigorously tie the phenomena we observe in practice to a misalignment between the (human-specified) notion of robustness and the inherent geometry of the data.