Built with and by Teycir Ben Soltane•
How to Use•FAQ•GitHub•arXiv.org•
Share:
ArXivCSExplorer
☆☆Bookmarks🏆RSSHow to UseFAQ
Back to Paper
cs.LGcs.AIcs.CR

Local ID: 2605.06505v2

AI Summary: gemma4:e4b

PACZero: PAC-Private Fine-Tuning of Language Models via Sign Quantization

By Murat Bilgehan Ertan, Xiaochen Zhu, Phuong Ha Nguyen, Marten van Dijk, Srinivas Devadas

Revision History Timeline

v15/7/2026
5/7/2026

No submitter comment provided.

v25/24/2026
5/24/2026

No submitter comment provided.

★ Version indexed in Explorer

Comparing v1 vs v2

Green = Added • Red = Removed

Title Comparison

PACZero: PAC-Private Fine-Tuning of Language Models via Sign Quantization

Authors Comparison

No author changes.

v1 Comment

No comment for this version.

v2 Comment

No comment for this version.

Abstract Word Diff

We introduce PACZero, a family of PAC-private zeroth-order mechanisms for fine-tuning large language models that delivers usable utility at $I(S^*; Y_{1:T})=0$. This privacy regime bounds the membership-inference attack (MIA) posterior success rate at the prior, an MIA-resistance level the DP framework matches only at $\varepsilon=0$ and infinite noise. All DP-ZO comparisons below are matched at the MIA posterior level. The key insight is that PAC Privacy charges mutual information only when the release depends on which candidate subset is the secret. Sign-quantizing subset-aggregated zeroth-order gradients creates frequent unanimity, steps at which every candidate subset agrees on the update direction; at these steps the released sign costs zero conditional mutual information. We propose two variants that span the privacy-utility trade-off: PACZero-MI (budgeted MI via exact calibration on the binary release) and PACZero-ZPL ($I=0$ via a uniform coin flip on disagreement steps). We evaluate on SST-2 and SQuAD with OPT-1.3B and OPT-6.7B in both LoRA and full-parameter tracks. On SST-2 OPT-1.3B full fine-tuning at $I=0$, PACZero-ZPL reaches ${88.99\pm0.91}$, within $2.1$pp of the non-private MeZO baseline ($91.1$ FT). No prior method produces usable utility in the high-privacy regime $\varepsilon<1$, and PACZero-ZPL obtains competitive SST-2 accuracy and nontrivial SQuAD F1 across OPT-1.3B and OPT-6.7B at $I=0$.
View Full Version History on arXiv