PulseAugur
实时 13:05:31
English(EN) SharedRequest: Privacy-Preserving Model-Agnostic Inference for Large Language Models

新方法增强了大型语言模型在提示、适应和 RAG 方面的隐私保护

研究人员开发了三种不同的方法来增强大型语言模型(LLM)的隐私保护。SharedRequest 提供了一个模型无关的框架,将提示与带噪声的变体混合,以在批处理级别隐藏敏感信息,从而提高可用性并降低推理成本。Echelon 提供了一个边界优先的训练架构,强制执行设备级别的模型状态非导出,从而实现跨隐私边界的可审计、仅聚合的适应。隐私感知解码(PAD)是一种轻量级的推理时防御机制,在生成过程中将校准后的噪声注入到令牌的 logits 中,专门用于检索增强生成(RAG)系统,以减轻私有信息泄露,同时保持响应的可用性。 AI

影响 这些进展为大型语言模型用户和开发者提供了更强的隐私保证,可能使其在敏感领域得到更广泛的应用。

排序理由 多篇研究论文提出了新的大型语言模型隐私保护方法。

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 4 个来源。 我们如何撰写摘要 →

新方法增强了大型语言模型在提示、适应和 RAG 方面的隐私保护

报道来源 [4]

  1. arXiv cs.AI TIER_1 English(EN) · Peihua Mai, Xuanrong Gao, Youlong Ding, Xianglong Du, Wei Liu, Yan Pang ·

    SharedRequest:面向大型语言模型的隐私保护模型无关推理

    arXiv:2606.05004v1 Announce Type: cross Abstract: With the widespread deployment of public large language models (LLMs) such as ChatGPT, protecting user prompt privacy has become an increasingly critical issue. Existing privacy-preserving inference methods sacrifice either utilit…

  2. arXiv cs.AI TIER_1 English(EN) · Yan Pang ·

    SharedRequest:面向大型语言模型的隐私保护、模型无关推理

    With the widespread deployment of public large language models (LLMs) such as ChatGPT, protecting user prompt privacy has become an increasingly critical issue. Existing privacy-preserving inference methods sacrifice either utility or efficiency, and often require model-specific …

  3. arXiv cs.AI TIER_1 English(EN) · Hina Dixit, Punit Kumar, Irene Tenison, Nevasini Sasikumar ·

    Echelon:跨越隐私边界的可审计聚合式语言模型适配

    arXiv:2606.02958v1 Announce Type: cross Abstract: Cross-organization language-model adaptation increasingly faces hard governance constraints: in many deployments, device-level model state-parameters, activations, optimizer state, and per-device updates-cannot be exported outside…

  4. arXiv cs.CL TIER_1 English(EN) · Haoran Wang, Xiongxiao Xu, Baixiang Huang, Kai Shu ·

    隐私感知解码:缓解检索增强生成中大型语言模型的隐私泄露

    arXiv:2508.03098v2 Announce Type: replace Abstract: Retrieval-Augmented Generation (RAG) enhances the factual accuracy of large language models (LLMs) by conditioning outputs on external knowledge sources. However, when retrieval involves private or sensitive data, RAG systems ar…