AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

TokenSwap: Backdoor Attack on the Compositional Understanding of Large Vision-Language Models

Zhifang Zhang, Qiqi Tao, Jiaqi Lv, Na Zhao, Lei Feng, Joey Tianyi Zhou

ICML 2026 · Adversarial robustness and security › Backdoor and poisoning · approach: Backdoors in large multimodal models · 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
add token swapping with adaptive token-weighted loss (TokenSwap vs BadNet)BLIP2 · Flickr30k (poisoned input)ASR (%): 53.13 -> 80.47Improved
add TokenSwap training instead of BadNetQwen-VL2.5-7B · Flickr30k (poisoned input)ASR (%): 45.35 -> 73.04Improved
add TokenSwap training instead of BadNetLLaVA-7B · Flickr30k (poisoned input)ASR (%): 78.91 -> 85.16Improved
add TokenSwap training instead of BadNetBLIP2 trained on MSCOCO · MSCOCO->Flickr8k (cross-dataset)ASR (%): 54.69 -> 88.28Improved
add TokenSwap loss instead of TrojVLM/MABA fixed-target attacksBLIP2 · Flickr8k (poisoned input, attack comparison)ASR (%): 80.47 -> 91.41Improved
add adaptive token-weighted emphasis (alpha > 0) (ablation)TokenSwap · Flickr8k ablation (alpha and gamma)ASR (%)Improved
add focal-style gamma weighting (ablation)TokenSwap · Flickr8k ablation (alpha and gamma)ASR (%)Improved
raise poisoning rate from 0.1 to 0.5 (ablation)TokenSwap · Flickr8k ablation (poisoning rate)ASR (%): 75.00 -> 91.41Improved
raise poisoning rate from 0.5 to 0.9 (ablation)TokenSwap · Flickr8k ablation (poisoning rate)ASR (%): 91.41 -> 92.97No clear effect
vary trigger type (Black to Blended noise) (ablation)TokenSwap · Flickr8k ablation (trigger type)ASR (%): 87.50 -> 92.19No clear effect
vary trigger size from 20 to 50 (ablation)TokenSwap · Flickr8k ablation (trigger size)ASR (%): 89.84 -> 90.63No clear effect
apply TokenSwap to VQA instead of BadNetBLIP2-style VQA models · VQAv2 (poisoned input)ASR (%): 79.89 -> 84.83Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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