Research Papers 论文研究 5h ago Updated 17m ago 更新于 17分钟前 48

Privacy Without Regret: Differentially Private Inference-Time Alignment 无悔隐私:差分隐私推理时对齐

Best-of-N (BoN) sampling, a widely used inference-time alignment strategy, suffers from reward hacking and lacks privacy protection for sensitive human preference data Adding calibrated Gumbel noise to reward scores before selection simultaneously achieves differential privacy and KL-regularized alignment, eliminating the need for separate regularization Private Best-of-N (PrivBoN) proves that when the privacy budget exceeds a critical threshold ε*, the privacy-mandated noise becomes regret-opti 提出在推理时向奖励分数添加校准噪声的单一干预,同时解决Best-of-N采样的奖励黑客攻击和隐私泄露问题 PrivBoN方法证明Gumbel噪声在适当尺度下可同时提供ε-差分隐私和KL正则化对齐,当隐私预算超过临界阈值ε*时隐私不增加额外对齐成本 PrivITP方法结合χ²正则化拒绝采样与两阶段高斯机制,实现ex-post (ε,δ)-DP,隐私成本与响应数量n无关 实验验证PrivBoN和PrivITP具有缩放单调性,PrivITP在强隐私场景下显著优于PrivBoN

62
Hot 热度
76
Quality 质量
70
Impact 影响力

Analysis 深度分析

TL;DR

  • Best-of-N (BoN) sampling, a widely used inference-time alignment strategy, suffers from reward hacking and lacks privacy protection for sensitive human preference data
  • Adding calibrated Gumbel noise to reward scores before selection simultaneously achieves differential privacy and KL-regularized alignment, eliminating the need for separate regularization
  • Private Best-of-N (PrivBoN) proves that when the privacy budget exceeds a critical threshold ε*, the privacy-mandated noise becomes regret-optimal regularization with zero additional alignment cost
  • Private Inference-Time Pessimism (PrivITP) combines χ²-regularized rejection sampling with a two-phase Gaussian mechanism, achieving ex-post (ε,δ)-DP with privacy cost independent of response count n
  • Experiments show both PrivBoN and PrivITP are scaling-monotonic (unlike BoN which degrades past a critical n), with PrivITP matching or outperforming PrivBoN, especially in strong-privacy regimes

Why It Matters

This work bridges two critical concerns in production LLM deployment—alignment quality and data privacy—by showing they can be addressed simultaneously through a single mechanism rather than requiring separate, potentially conflicting solutions. For AI practitioners deploying reward-model-based alignment at inference time, this provides a theoretically grounded, privacy-preserving alternative that avoids the well-known reward hacking problem while protecting the sensitive human preference data underlying reward model training.

Technical Details

  • PrivBoN: Adds Gumbel noise at a calibrated scale to reward scores before Best-of-N selection. This noise simultaneously provides ε-differential privacy and implements KL-regularized alignment, with the privacy-mandated noise becoming regret-optimal when ε > ε*
  • PrivITP: A more robust method combining χ²-regularized rejection sampling with a two-phase Gaussian mechanism, achieving ex-post (ε,δ)-differential privacy. It cleanly decouples the regularization parameter from the privacy parameter
  • Critical threshold ε*: Depends on an unknown coverage coefficient, motivating the introduction of PrivITP as a practical alternative that does not require this knowledge
  • Scaling behavior: Unlike standard BoN which degrades beyond a critical number of sampled responses n, both PrivBoN and PrivITP are scaling-monotonic, maintaining or improving performance as n increases
  • Experimental validation: Tested across multiple language models, datasets, and reward models, confirming PrivITP matches or exceeds PrivBoN performance at equivalent privacy levels, with largest gains in strong-privacy regimes

Industry Insight

  • The theoretical equivalence between differential privacy and KL-regularization suggests that privacy-preserving inference-time alignment may become a standard requirement as regulatory frameworks tighten around training data provenance, making these techniques increasingly relevant for production deployments
  • The scaling-monotonic property of PrivBoN and PrivITP addresses a practical pain point: practitioners can safely increase N for better alignment without the degradation risk inherent in standard BoN, simplifying hyperparameter tuning
  • The decoupling of privacy and regularization parameters in PrivITP offers operational flexibility, allowing teams to independently tune privacy guarantees and alignment quality—a significant advantage for organizations navigating evolving compliance requirements

TL;DR

  • 提出在推理时向奖励分数添加校准噪声的单一干预,同时解决Best-of-N采样的奖励黑客攻击和隐私泄露问题
  • PrivBoN方法证明Gumbel噪声在适当尺度下可同时提供ε-差分隐私和KL正则化对齐,当隐私预算超过临界阈值ε*时隐私不增加额外对齐成本
  • PrivITP方法结合χ²正则化拒绝采样与两阶段高斯机制,实现ex-post (ε,δ)-DP,隐私成本与响应数量n无关
  • 实验验证PrivBoN和PrivITP具有缩放单调性,PrivITP在强隐私场景下显著优于PrivBoN

为什么值得看

本文首次将差分隐私与推理时对齐策略结合,为LLM部署提供了隐私保护的理论保证和实践方案。研究揭示了隐私保护与模型对齐之间的内在联系,为工业界在敏感场景下部署AI系统提供了重要参考。

技术解析

  • PrivBoN方法:在奖励模型评分阶段注入Gumbel噪声,噪声尺度由隐私预算ε控制。理论证明当ε > ε*时,隐私强制的噪声恰好实现 regret-optimal 的KL正则化对齐,达到信息论上界。
  • PrivITP方法:采用两阶段高斯机制,第一阶段对χ²正则化拒绝采样添加噪声,第二阶段进行隐私聚合。关键创新在于隐私成本与采样数量n解耦,避免了传统方法随n增大隐私预算快速耗尽的问题。
  • 实验设置:在多个语言模型、数据集和奖励模型上进行验证,对比了BoN、PrivBoN和PrivITP三种方法在不同隐私预算下的性能表现。

行业启示

  • 推理时对齐成为隐私保护的新范式:无需修改训练流程,在部署阶段即可实现隐私保护,降低了企业落地差分隐私AI系统的门槛。
  • 隐私与性能可兼得:研究证明在足够隐私预算下,隐私保护不增加额外对齐成本,打破了"隐私必牺牲性能"的传统认知。
  • 建议企业在敏感应用场景(如医疗、金融)优先采用PrivITP等推理时隐私保护方案,同时关注隐私预算ε的合理设置以平衡隐私与效用。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

LLM 大模型 Alignment 对齐 Security 安全 Inference 推理 Research 科学研究