.png){kind=icon}
Return
Generating transferable attacks across large vision-language models using adversarial deformation learning
.png){kind=icon}
PRE
AI
DOI：10.1016/j.patcog.2026.113194.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• This paper develops a novel cross-model attack method against LVLMs via adversarial deformation learning. • Novel cross-modal consistent perturbations are developed through adversarial learning to improve the transferability. • Prompt purification and adversarial transformation networks are introduced to simulate potential deformations within unseen LVLMs. • Extensive experiments are conducted on various LVLM models and datasets to illustrate both the effectiveness and transferability of our method.
Journal
IF:
7.6
Papers:
1.3W
Citations:
4.5W
Organization
Cited Papers
Prompting large language model with context and pre-answer for knowledge-based VQA.png)
PATTERN RECOGNITION
PATTERN RECOGNITIONIF7.6
PREAI
ShareSaveSentiment analysis based on text information enhancement and multimodal feature fusionPATTERN RECOGNITION
PATTERN RECOGNITIONIF7.6
PREAI
ShareSave
All-in-One Weather-Degraded Image Restoration Via Adaptive Degradation-Aware Self-Prompting Model
PREAI
ShareSaveShareSaveShareSave
View more