AI Security

To address the threat of model tampering in deep learning models, the study proposes a method based on a fragile watermarking scheme to accurately locate and recover tampered model parameters.