arXiv cs.CL· Zihao Zhao, Avi Schwarzschild, Casey Meehan, Katherine Lee, Zhili Feng, Dylan Sam·· 12 小时前AI 评分31
合成医疗记录的隐私实证研究：隐私感知改写优于 DP-SGD 与直接微调
Exploring The Empirical Privacy Landscape of Synthetic Data
AI 导读
一项受控对比研究在带 canary 探针的合成医疗记录上比较了数据级与训练级干预的能力-隐私权衡：直接微调能力指标最好但隐私泄漏最多，DP-SGD 显著降低泄漏却以能力大幅下降为代价。用唯一性感知指令引导生成合成数据、以隐私感知改写替代朴素复述，可在多种改写模型上提升隐私保护。能力更强的改写模型通常降低下游隐私泄漏风险，但留出集拟合并非单调改善，增加测试时计算带来的收益也不稳定。
正文 · 原文
Abstract:Methods for training language models on sensitive data differ in how they balance capability gains against the risk of leaking private information. In this paper, we present a controlled comparison of data-level and training-level interventions for navigating the capability-privacy tradeoff on synthetic medical records with canaries to probe for leakage. Direct finetuning performs best on capability metrics but leads to the most privacy leakage, while DP-SGD reduces leakage at substantial cost to capabilities. Generating synthetic data from these records offers a more favorable balance, particularly when guided by a uniqueness-aware instruction to replace private details. Relative to naive paraphrasing, privacy-aware rewriting improves privacy protection across a variety of different rewriting models. We also find that more capable rewriting models generally reduce the risks of downstream privacy leakage, although held-out fit does not improve monotonically and increased test-time compute yields inconsistent gains. Overall, these results highlight privacy-aware rewriting as a promising direction for achieving a more favorable trade-off between capabilities and empirical privacy compared to other training-level methods, and our evaluation framework supports comparisons of future models and methods.
| Subjects: | Cryptography and Security (cs.CR); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.22659 [cs.CR] |
| (or arXiv:2610.22659v1 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2610.22659 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Zihao Zhao [view email]
[v1]
Thu, 15 Oct 2026 21:22:35 UTC (560 KB)
来源：arXiv cs.CL · arxiv.org