通过批量混噪提示，实现无需修改模型的高效隐私保护推理
SharedRequest: Privacy-Preserving Model-Agnostic Inference for Large Language Models
- 将隐私保护从单个请求升级为批量处理，用噪声提示混合掩盖敏感信息
- 相比差分隐私基线，响应质量提升超20%，查询成本降低至五分之一
- 不依赖模型结构，适用于任意大语言模型，适合注重隐私与效率的部署
随着ChatGPT等公开大语言模型的广泛部署，用户提示隐私保护日益重要。现有隐私保护推理方法或牺牲性能，或需特定模型修改，兼容性受限。本文提出SharedRequest，一种模型无关的隐私保护大语言模型推理框架，将隐私保护从单个提示层面转移到批量层面。核心思想是通过混合原始提示与噪声变体来模糊敏感信息，同时将语义相似指令分组，使推理开销在大批量请求中分摊，对模型输出质量影响极小。该设计不依赖模型架构，无需访问参数或修改结构。实验表明，SharedRequest在保持高隐私性的同时，相比差分隐私基线提升了超过20%的实用性，共享提示机制使查询成本相比非批处理推理降低达5倍。
原文摘要 · Abstract (English)
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 modifications that limit their compatibility. In this paper, we propose SharedRequest, a model-agnostic framework for privacy-preserving LLM inference that reformulates privacy protection at the batch level rather than the individual-prompt level. The key idea is to obscure sensitive information by mixing original prompts with noisy variants, while grouping semantically equivalent instructions to amortize the inference cost over a large batch of queries with minimal impact on LLM response quality. This design is independent of the LLM architecture, requiring no access to model parameters or architectural modification. Empirical results demonstrate that SharedRequest achieves over $20\%$ higher utility compared to prior differential privacy baselines, and its shared-prompt mechanism reduces query cost by up to $5\times$ compared to non-batched inference.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。