1 paper · 1 filter
Jun Liu, Leo Yu Zhang, Fengpeng Li +2
Hard-label black-box attacks, relying solely on top-1 predictions, represent one of the most challenging yet practically threat models. Despite recent progress, existing approaches…