adversarial attack detection, model fine-tuning, image pyramid, information entropy redundancy, multi-scale feature verification, heterogeneous model collaboration
Large language models, Jailbreak attacks, Safety alignment, Latent representation, Manifold projection, Adversarial robustness
