Research Papers 论文研究 1d ago Updated 10h ago 更新于 10小时前 35

PAC-Private Autoregressive Generation: Calibrating Noise to Ensemble Disagreement PAC-Private Autoregressive Generation: Calibrating Noise to Ensemble Disagreement

Extends PAC (Private Autocorrelation) privacy from classification to autoregressive text generation, a first in the literature Uses 128 overlapping "worlds" from the private corpus, each training a separate adapter over a frozen public model, with noise calibrated to ensemble disagreement rather than fixed per-token budgets Introduces coupled decoding to prevent greedy degeneration while preserving privacy accounting under adaptive self-generated contexts Achieves 74% retention of fine-tuning ga 首次将PAC隐私框架从分类任务扩展至自回归文本生成,解决API服务中私有文本的隐私泄露问题 构建128个重叠世界,每个记录出现在64个世界中,训练对应adapter,通过世界投票的离散度动态校准噪声 在WikiText-103/GPT-2-small上,以2^-32每token预算保留74%微调增益,百万token后成员推断成功率≤51.08% 相比PMixED方法,在相同隐私预算下保留98%非隐私性能余量(PMixED仅56%),且无性能交叉点 提出耦合解码机制,在保持隐私计量的同时避免贪婪解码退化,并证明互信息上界I(S;Y_{1:T})≤bT

50
Hot 热度
50
Quality 质量
50
Impact 影响力

Analysis 深度分析

TL;DR

  • Extends PAC (Private Autocorrelation) privacy from classification to autoregressive text generation, a first in the literature
  • Uses 128 overlapping "worlds" from the private corpus, each training a separate adapter over a frozen public model, with noise calibrated to ensemble disagreement rather than fixed per-token budgets
  • Introduces coupled decoding to prevent greedy degeneration while preserving privacy accounting under adaptive self-generated contexts
  • Achieves 74% retention of fine-tuning gain at a per-token budget of 2^-32 on WikiText-103 with GPT-2-small, with membership-inference success bounded at 51.08% after 10^6 tokens
  • Outperforms PMixED significantly, retaining 98% of non-private headroom versus at most 56% across 10^2 to 10^6 tokens under matched membership-inference bounds

Why It Matters

This work addresses a critical gap in privacy-preserving AI: while private inference methods exist for classification, autoregressive generation—where models serve as APIs producing sequential outputs—remains largely unprotected. For practitioners deploying fine-tuned language models on sensitive corpora, this framework offers a practical path to limit privacy leakage through generated text without catastrophic utility loss, making it directly relevant to any organization serving private-data-adapted LLMs via API.

Technical Details

  • World construction: 128 overlapping worlds are built from the private corpus, with each record appearing in exactly 64 worlds; one LoRA adapter is trained per world over a frozen public model (GPT-2-small), and the realized world constitutes the secret
  • Noise calibration mechanism: At each token step, the public model defines a candidate set, the 128 world-adapters vote, and posterior-weighted disagreement across worlds determines the magnitude of PAC noise added; unanimous predictions require zero calibration noise
  • Theoretical guarantee: Proves that mutual information between the secret S and generated output Y_{1:T} is bounded by I(S;H_T) ≤ bT, where H_T represents the hidden state at step T
  • Coupled decoding: A novel decoding strategy that preserves privacy accounting while avoiding the degeneration that occurs with naive greedy decoding under noise injection
  • Empirical evaluation: Tested on WikiText-103 with GPT-2-small; posterior-entropy estimates of actual leakage are ~17% of the charged privacy budget, indicating substantial margin in the theoretical bounds

Industry Insight

  • The finding that inference privacy does not equal content protection—memorized canaries are emitted at the same rate even when membership advantage is near zero—should caution practitioners against over-relying on membership-inference metrics as the sole privacy guarantee; complementary content-level safeguards remain necessary
  • The dramatic performance gap versus PMixED (98% vs 56% headroom retention) suggests that fixed-per-token privacy accounting is fundamentally ill-suited for long-horizon autoregressive tasks, and adaptive noise calibration should become a standard design principle
  • The adapter-per-world architecture, while elegant, scales linearly with the number of worlds; for production deployment with large corpora, researchers should explore compressed world representations or sparse ensemble selection to reduce computational overhead without sacrificing privacy guarantees

TL;DR

  • 首次将PAC隐私框架从分类任务扩展至自回归文本生成,解决API服务中私有文本的隐私泄露问题
  • 构建128个重叠世界,每个记录出现在64个世界中,训练对应adapter,通过世界投票的离散度动态校准噪声
  • 在WikiText-103/GPT-2-small上,以2^-32每token预算保留74%微调增益,百万token后成员推断成功率≤51.08%
  • 相比PMixED方法,在相同隐私预算下保留98%非隐私性能余量(PMixED仅56%),且无性能交叉点
  • 提出耦合解码机制,在保持隐私计量的同时避免贪婪解码退化,并证明互信息上界I(S;Y_{1:T})≤bT

为什么值得看

该工作填补了PAC隐私在自回归生成领域的空白,为私有化LLM的API服务提供了可证明的隐私保障方案。其动态噪声校准策略显著优于传统方法,在严格隐私约束下仍保持接近非隐私模型的性能,对隐私敏感场景(如企业知识库、个人数据)的模型部署具有重要参考价值。

技术解析

  • 多世界架构:从私有语料构建m=128个重叠世界,每个记录恰好出现在m/2=64个世界中,冻结公共模型(GPT-2-small)并在每个世界训练独立adapter,实现世界作为隐私秘密的编码
  • 动态噪声校准:在每token生成时,公共模型定义候选集,各世界投票后通过后验加权离散度确定PAC噪声量;当所有世界意见一致时不加校准噪声,实现隐私-效用自适应平衡
  • 耦合解码机制:引入耦合解码策略,在保持隐私预算累加可追踪的同时,避免传统噪声注入导致的贪婪解码退化问题,确保生成品质
  • 理论保证:证明互信息上界I(S;Y_{1:T})≤I(S;H_T)≤bT,其中S为秘密(实现世界),Y_{1:T}为生成序列,H_T为历史,b为每步隐私预算
  • 实验基准:在WikiText-103上评估,后验熵估计的泄露量仅为收费预算的约17%,表明实际隐私保护远优于理论最坏情况

行业启示

  • 私有化LLM服务不应仅依赖权重保护,需建立输出端的隐私计量体系;PAC框架的动态校准思路可推广至其他生成任务(如图像、代码生成)
  • 企业部署私有模型API时,建议采用多adapter集成架构替代单一微调模型,在隐私预算可控前提下最大化性能保留率
  • 当前隐私保护仍不等于内容保护(如记忆canary仍以相同速率输出),需结合内容过滤、差分隐私等复合策略构建纵深防御体系

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。