English(EN) UNDREAM: Bridging Differentiable Rendering and Photorealistic Simulation for End-to-end Adversarial Attacks

新的UNDREAM框架可在模拟中实现端到端的对抗性攻击

作者 PulseAugur 编辑部 · [1 个来源] · 2026-07-07 04:00

研究人员开发了UNDREAM，一个旨在弥合照片级真实感模拟器与可微分渲染器之间差距的新型软件框架。这项创新实现了在模拟环境中对3D对象上的对抗性扰动的端到端优化。UNDREAM提供了对天气、光照和相机角度等环境因素的全面控制，有助于创建多样化且逼真的对抗性攻击场景。该框架将照片级真实感模拟与可微分优化相结合的能力，有望推动物理对抗性攻击的研究。 AI

影响能够对安全关键应用中AI的鲁棒性进行更真实、更全面的对抗性攻击测试。

排序理由关于新的对抗性攻击软件框架的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

Mansi Phute

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

报道来源 [1]

arXiv cs.AI TIER_1 English(EN) · Mansi Phute, Matthew Hull, Haoran Wang, Alec Helbling, ShengYun Peng, Willian Lunardi, Martin Andreoni, Wenke Lee, Duen Horng Chau · 2026-07-07 04:00

UNDREAM: Bridging Differentiable Rendering and Photorealistic Simulation for End-to-end Adversarial Attacks

arXiv:2510.16923v3 Announce Type: replace-cross Abstract: Deep learning models deployed in safety critical applications like autonomous driving use simulations to test their robustness against adversarial attacks in realistic conditions. However, these simulations are non-differe…

报道来源 [1]

UNDREAM: Bridging Differentiable Rendering and Photorealistic Simulation for End-to-end Adversarial Attacks

相关话题