AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

THE SELF-RE-WATERMARKING TRAP: FROM EXPLOIT TO RESILIENCE

Vithurabiman Senthuran, Yong Xiang, Iynkaran Natgunanathan, Uthayasanker Thayasivam

ICLR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Neural network watermarking

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add re-watermarking adversarial training and spectral normalization (full model) vs existing SOTAself-re-watermarking attack · COCOACC_orig (%): 49.90 -> 100.00Improved
add spectral normalization (Lipschitz constraint) (ablation)Proposed · COCOACC_orig after self overwrite (%): 76.33 -> 100.0Improved
add spectral normalization (ablation)Proposed · COCOPSNR (dB): 30.40 -> 34.03Improved
add spectral normalization (ablation)Proposed · COCOACC_orig after strong PGD (%): 98.90 -> 99.37Improved
add post-processing module (ablation)Proposed · COCOPSNR (dB): 31.82 -> 34.03Improved
add post-processing module (ablation)Proposed · COCOACC_orig after strong PGD (%): 99.99 -> 99.37No clear effect
increase lambda_LPIPS from 0.5 to 0.7 (ablation)Proposed · COCOPSNR (dB) / JPEG(50) ACC: 34.03 -> 39.51; 95.06 -> 88.08Mixed
decrease lambda_LPIPS from 0.5 to 0.3 (ablation)Proposed · COCOPSNR (dB) / JPEG(50) ACC: 34.03 -> 32.39; 95.06 -> 99.63Mixed
increase removal attack step size before re-embeddingProposed · COCOACC_orig (%): 68.11 -> 53.00Hurt
train on omitting robustness loss in adversary model (AM) vs BAMadversary overwriting model · MIRFLICKRACC_orig of attacker's output: 1.00 vs 0.99No clear effect
increase PGD perturbation budget epsilon above 0.05Proposed · COCObit accuracyHurt
train and test on other datasets (MIRFLICKR, CelebA, ImageNet)Proposed · MIRFLICKRACC_orig self re-embed (%): 100.00 -> 100.00No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy