Checkerboard attack offers efficient, learning-free backdoor for deep learning models

Researchers have developed a new method called Checkerboard for launching clean-label backdoor attacks on deep learning models. This learning-free technique uses a closed-form checkerboard trigger derived from linear separability, eliminating the need for complex training or optimization. Checkerboard has demonstrated high effectiveness and efficiency across benchmark datasets like CIFAR-10 and ImageNet-100, even with very low poisoning budgets and against advanced defenses. AI

影响 Introduces a new, efficient method for backdoor attacks that could challenge the security of deployed deep learning models.

排序理由 This is a research paper detailing a novel attack method on deep learning models. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CV 阅读 →

• Checkerboard
• CIFAR-10
• ImageNet-100

• 论文
• 安全

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