AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

On the Vulnerability of Parameter-Level Defenses to Model Merging

Kuangpu Guo, Jian Liang, Qingyan Zheng, Yu Yongcan, Zilei Wang, Ran He, Tieniu Tan

ECCV 2026 · Adversarial robustness and security › Backdoor and poisoning · approach: Detecting poisoned samples · 0 citations (OpenAlex)

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
apply AGA attack to MergeLock-protected modelsTask Arithmetic merge · ViT-B/32 protected taskavg accuracy: 5.10 -> 66.39Improved
apply AGA attack to MergeLock-protected modelsLOT Merging · ViT-B/32 protected taskavg accuracy: 4.85 -> 78.36Improved
apply AGA attack to MergeBarrier-protected modelsTask Arithmetic merge · ViT-B/32 protected taskavg accuracy: 29.32 -> 60.80Improved
apply AGA attack to Params-D-protected modelsQwen2-7B · Alpaca/GSM8K/MBPPavg score: 3.07 -> 47.79Improved
apply AGA attack to MergeLock-protected modelsViT-L/14 · protected taskavg accuracy: 4.27 -> 82.72Improved
combine Params, MergeLock and MergeBarrier defenses then attack with AGAcombined defense · ViT-B/32 protected taskavg accuracy: 4.16 -> 60.66Improved
apply AGA to MergeLock-protected merged modelTask Arithmetic merge · cross-task average utilityavg accuracy: 4.92 -> 67.17Improved
replace MergeGuard with ARF protectionARF · ViT-B/32 protected task under AGAavg accuracy: 50.56 -> 27.42Improved
replace MergeGuard with ARF protectionARF · ViT-L/14 protected task under AGAavg accuracy: 63.60 -> 29.26Improved
replace MergeGuard with ARF protectionARF · ViT-B/32 standalone accuracyavg accuracy: 86.31 -> 88.11Improved
increase ARF expansion ratio rho from 0.02 to 0.08 (ablation)ARF · ViT-B/32 protected under AGAavg accuracy: 28.75 -> 26.36Mixed
increase ARF lambda_dist from 0.7 to 1.3 (ablation)ARF · ViT-B/32 protected under AGAavg accuracy: 29.26 -> 26.17Mixed

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