English(EN) Z-PEFT: Zero-shot Backdoor Detection in Parameter-Efficient Fine-Tuning via Canonical Spectral Signatures

新的Z-PEFT方法可检测微调AI模型中的后门

作者 PulseAugur 编辑部 · [2 个来源] · 2026-08-03 14:10

研究人员开发了Z-PEFT，一种用于检测参数高效微调（PEFT）模型中后门的新颖方法。该方法利用模型权重的规范谱签名来识别恶意模型，即使面对未见过的攻击或数据集。Z-PEFT为增强广泛共享的PEFT模型的安全性提供了一种轻量级且计算高效的解决方案。 AI

影响通过提供一种强大的方法来检测恶意微调，增强了共享AI模型的安全性。

排序理由该集群描述了一篇研究论文，详细介绍了一种检测AI模型中后门的新方法。

AI 生成摘要 · Google Gemini · 来自 2 个来源。我们如何撰写摘要 →

报道来源 [2]

arXiv cs.LG TIER_1 English(EN) · Nicola Pitzalis, Donald Shenaj, Giacomo Cignoni, Andrea Cossu, Davide Bacciu, Antonio Carta · 2026-08-04 04:00

Z-PEFT: Zero-shot Backdoor Detection in Parameter-Efficient Fine-Tuning via Canonical Spectral Signatures

arXiv:2608.02271v1 Announce Type: new Abstract: Parameter-Efficient Fine-tuned (PEFT) models are frequently downloaded from open repositories by practitioners. This widespread practice creates a significant attack surface, as malicious actors can publish backdoored models that in…
Hugging Face Daily Papers TIER_1 English(EN) · 2026-08-03 14:10

Z-PEFT: Zero-shot Backdoor Detection in Parameter-Efficient Fine-Tuning via Canonical Spectral Signatures

Parameter-Efficient Fine-tuned (PEFT) models are frequently downloaded from open repositories by practitioners. This widespread practice creates a significant attack surface, as malicious actors can publish backdoored models that induce specific behaviors in response to predefine…