Disentangling Safe and Unsafe Corruptions via Anisotropy and Locality
This addresses the vulnerability of ML systems to small perturbations by providing a more realistic and adaptable threat model for practitioners in computer vision and classification tasks, though it is incremental in refining threat modeling rather than introducing a new paradigm.
The paper tackles the problem of defining threat models for adversarial robustness beyond isotropic and global norms by proposing a novel threat model called Projected Displacement (PD) that measures threat based on alignment with unsafe directions identified locally from training data. Experiments on Imagenet-1k show that PD includes safe perturbations like noise and blur while excluding unsafe ones, offering a flexible, task-driven specification without requiring pre-training.
State-of-the-art machine learning systems are vulnerable to small perturbations to their input, where ``small'' is defined according to a threat model that assigns a positive threat to each perturbation. Most prior works define a task-agnostic, isotropic, and global threat, like the $\ell_p$ norm, where the magnitude of the perturbation fully determines the degree of the threat and neither the direction of the attack nor its position in space matter. However, common corruptions in computer vision, such as blur, compression, or occlusions, are not well captured by such threat models. This paper proposes a novel threat model called \texttt{Projected Displacement} (PD) to study robustness beyond existing isotropic and global threat models. The proposed threat model measures the threat of a perturbation via its alignment with \textit{unsafe directions}, defined as directions in the input space along which a perturbation of sufficient magnitude changes the ground truth class label. Unsafe directions are identified locally for each input based on observed training data. In this way, the PD threat model exhibits anisotropy and locality. Experiments on Imagenet-1k data indicate that, for any input, the set of perturbations with small PD threat includes \textit{safe} perturbations of large $\ell_p$ norm that preserve the true label, such as noise, blur and compression, while simultaneously excluding \textit{unsafe} perturbations that alter the true label. Unlike perceptual threat models based on embeddings of large-vision models, the PD threat model can be readily computed for arbitrary classification tasks without pre-training or finetuning. Further additional task annotation such as sensitivity to image regions or concept hierarchies can be easily integrated into the assessment of threat and thus the PD threat model presents practitioners with a flexible, task-driven threat specification.