AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

From Internal Diagnosis to External Auditing: A VLM-Driven Paradigm for Data-Free Online Backdoor Defense

Binyan Xu, Fan Yang, Xilin Dai, Di Tang, Kehuan Zhang

ICML 2026 · Adversarial robustness and security › Backdoor and poisoning · approach: Detecting poisoned samples

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
use PRISM external VLM auditing instead of internal-diagnosis defensesdata-free backdoor defenses · CIFAR-10 (average over 11 attacks)ASR: Refine 14.3 -> PRISM 0.8Improved
use PRISM instead of zero-shot VLM classificationVLM auditor · CIFAR-10 (average over 11 attacks)clean accuracy: 86.7 -> 93.2Improved
average victim and VLM logits (Ensemble) instead of PRISM routingvictim + VLM · CIFAR-10 (average over 11 attacks)ASR: 0.8 -> 75.7Hurt
add online update to PRISM (ablation)PRISM · GTSRB, classic attacksASR: 32.9 -> 1.3Improved
add prototype refinement to PRISM (ablation)PRISM · GTSRB, classic attacksclean accuracy: 86.1 -> 94.4Improved
add Cornish-Fisher skewness correction to the adaptive threshold (ablation)PRISM · GTSRB, classic attacksASR and CA: ASR 0.8 (w/o) vs 1.3 (full)No clear effect
add skewness correction so a single threshold zeta=-2 works across datasets (ablation)PRISM · CIFAR and GTSRB threshold sensitivityoptimal threshold consistencyImproved
typographic adaptive prompt attack superimposing target class name (ablation)PRISM · CIFAR-10ASR: 1.3 -> 15.6Hurt
backdoor the auditing VLM with the same trigger and target as the victim (ablation)PRISM · CIFAR-10, BadNetsASR: 0.2 -> 100Hurt
backdoor the auditing VLM with a different trigger or different target (ablation)PRISM · CIFAR-10, BadNetsASR: 0.2 -> 0.8 (different trigger), 0.0 (different target)No clear effect
flooding, periodic and mixed online adaptive attacks (ablation)PRISM · GTSRB online streamASRNo clear effect
use poison set 50x larger than clean set (ablation)PRISM · dataset size scalabilityASR: ~12%Hurt

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