AI Paper Atlas

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

AdaShield: Safeguarding Multimodal Large Language Models from Structure-based Attack via Adaptive Shield Prompting

Yu Wang, Xiaogeng Liu, Yu Li, Muhao Chen, Chaowei Xiao

ECCV 2024 · Adversarial robustness and security › Attacks on text and graphs · approach: LLM jailbreak and injection defense · 77 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
prepend adaptive auto-refined defense prompt (AdaShield-A) vs vanillaLLaVA 1.5-13B · QR attackASR: 75.75 -> 15.22Improved
prepend AdaShield-A defense prompt vs vanillaCogVLM-chat-v1.1 · FigStep attackASR: 85.19 -> 0.00Improved
use AdaShield-A instead of static prompt AdaShield-S (ablation)LLaVA 1.5-13B · QR attackASR: 24.43 -> 15.22Improved
add post-hoc harm detector and detoxifier (MLLMP) vs vanillaLLaVA 1.5-13B · QR attackASR: 75.75 -> 77.96No clear effect
static defense prompt FSD on benign tasksCogVLM-chat-v1.1 · MM-Vet benignTotal: 50.0 -> 27.4Hurt
AdaShield-S on MiniGPT-v2 benign tasks (over-defense)MiniGPT-v2-13B · MM-Vet benignTotal: 14.8 -> 1.4Hurt
retrieve defense prompts by similarity (threshold) instead of random choice (ablation)AdaShield-A · QR attack, LLaVA 1.5-13BASR: 18.20 -> 15.22Improved
add instruction to execute benign queries safely (intuition 4) (ablation)AdaShield-S · MM-Vet, CogVLM-chat-v1.1Total: 14.3 -> 45.9Improved
specify explicit refusal response ('I am sorry') in defense prompt (ablation)AdaShield-S · QR and FigStepASRImproved
train defense pool on Hard scenarios only vs All (ablation)AdaShield-A · QR (All test)ASR: 15.22 -> 14.63Improved
transfer defense prompts learned with another MLLM (ablation)AdaShield-A · QR, CogVLM-chat-v1.1ASR: 16.07 -> 7.33Improved

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