Safety
47d agoNew Framework Certifies MLP Adversarial Robustness
A new theoretical framework reduces the adversarial robustness problem to a lattice traversal problem. Researchers use axis-aligned hyper-rectangles to define sound and complete certifications for multilayered perceptrons. This approach provides a rigorous way to prove that input perturbations won't change a classifier's prediction.
The Signal
It offers a formal verification tool for safety-critical MLPs.