Ramchandran Muthukumar

Disentangling Safe and Unsafe Corruptions via Anisotropy and Locality

2025 · arXiv (Cornell University)

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, …