✨Toward the Frontiers of Reliable Diffusion Sampling via Adversarial Sinkhorn Attention Guidance

📝 Summary:
ASAG is a novel diffusion guidance method that uses optimal transport and the Sinkhorn algorithm to adversarially disrupt attention scores. It weakens misleading attention alignments by injecting an adversarial cost, improving sample quality, controllability, and fidelity without model retraining.

🔹 Publication Date: Published on Nov 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.07499
• PDF: https://arxiv.org/pdf/2511.07499

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#DiffusionModels #AdversarialAI #OptimalTransport #GenerativeAI #DeepLearning

✨Few Tokens Matter: Entropy Guided Attacks on Vision-Language Models

📝 Summary:
Targeting high-entropy tokens in vision-language models causes significant semantic degradation with reduced budgets. This attack strategy reveals critical transferable safety risks across different VLM architectures.

🔹 Publication Date: Published on Dec 26, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.21815
• PDF: https://arxiv.org/pdf/2512.21815

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#VisionLanguageModels #AdversarialAI #AIsecurity #MachineLearning #DeepLearning

✨Few Tokens Matter: Entropy Guided Attacks on Vision-Language Models

📝 Summary:
Attacking a few high-entropy tokens in VLMs significantly degrades outputs with reduced budgets. These selective attacks efficiently create harmful outputs and transfer across architectures, exposing new VLM safety weaknesses.

🔹 Publication Date: Published on Dec 26, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.21815
• PDF: https://arxiv.org/pdf/2512.21815

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#VLMs #AISafety #AdversarialAI #MachineLearning #AIResearch

✨Tex3D: Objects as Attack Surfaces via Adversarial 3D Textures for Vision-Language-Action Models

📝 Summary:
Tex3D is the first framework optimizing 3D adversarial textures to attack vision-language-action models. It significantly degrades robotic manipulation performance in real-world settings, revealing critical vulnerabilities.

🔹 Publication Date: Published on Apr 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.01618
• PDF: https://arxiv.org/pdf/2604.01618
• Project Page: https://vla-attack.github.io/tex3d/
• Github: https://github.com/vla-attack/tex3d

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#AdversarialAI #Robotics #VLAmodels #Cybersecurity #ComputerVision