New framework discovers natural transformation vulnerabilities in vision models

Researchers have developed a new framework called Adversarial Scenario Attack (ASA) to identify vulnerabilities in black-box vision models. ASA utilizes a multimodal language model and a text-guided generative editor to explore various natural transformations, such as changes in background, weather, and materials. This method has demonstrated higher attack success rates on ImageNet classifiers compared to previous query-based generative attacks, while requiring fewer queries and maintaining perceptual quality. AI

IMPACT This research could lead to more robust vision models by identifying and addressing their weaknesses to realistic environmental changes.

RANK_REASON The cluster contains a research paper detailing a new method for discovering vulnerabilities in AI models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Adversarial Scenario Attack (ASA)
• alphaXiv
• arXiv
• CatalyzeX
• Connected Papers
• DagsHub
• Gotit.pub
• Hugging Face
• ImageNet
• Litmaps
• ScienceCast

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →