Pith. sign in

REVIEW 3 major objections 5 minor 24 references

Hush! Protecting Secrets During Model Training: An Indistinguishability Approach

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read A new per-secret protection definition and an LP-based sampling algorithm that trains models with substantially lower noise than DP-SGD while bounding the posterior probability of secret reconstruction.

desk verdict A genuinely new per-secret protection definition and a promising LP-based training pipeline, but the current version has a likely factor-of-two error in the privacy calibration and a dubious scaling assumption that undercut the empirical claims. read the letter →

arxiv 2506.00201 v1 pith:DVNHGSYI submitted 2025-05-30 cs.CR

classification cs.CR
keywords secretdatamodelalgorithmdefinitionprotectiontrainingdataset
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper studies a company that wants to train a model on its own private data without letting the model reveal specific secrets inside that data. Standard differential privacy (DP) is very strong, but it costs a lot of accuracy and assumes the data owner is different from the data user. Here, the same organization owns the data and the model, so the authors propose a weaker, more targeted goal: for each secret, guarantee that after seeing the model, an adversary's chance of reconstructing that secret is below a chosen limit, starting from a chosen prior.

The method has two parts. First, compute for each secret a privacy budget number mu, derived from the desired prior and posterior probabilities. Second, solve a linear program that assigns each training example a weight, so that total weight assigned to each secret is capped by its budget, while keeping as much data as possible. Examples are then sampled with probabilities proportional to their weights, and DP-SGD is run with Poisson sampling and a noise level calibrated to the actual sampling probabilities. On an arXiv abstract fine-tuning task, the preprocessing reduces the required noise by about 8x and improves test loss by about 7.5% compared to running DP-SGD on the whole dataset.

The proof backbone is an f-divergence generalization of Fano's inequality (Lemma 3.2) that translates a KL divergence bound into a reconstruction-probability bound. A separate experiment pretrains Gemma-style models with small Gaussian noise and shows that extraction of training text drops sharply, supporting the idea that moderate indistinguishability can hinder reconstruction.

Extended reading notes

Core claim

The paper's central claim is that for a same-owner secret-protection setting, bounding posterior reconstruction probability per secret (Definition 3.1) can be achieved by LP-weighted Poisson sampling plus DP-SGD with KL-calibrated noise, and that this yields reductions of about 8x in noise multiplier and about 7.5% in test loss versus running DP-SGD on the full dataset. If correct, the formal guarantee is the stated bound on reconstruction probability for each secret.

Load-bearing premise

The load-bearing assumption is that the data owner can reliably specify the secret set S, the containment function E (which examples contain which secrets), and the transformation predicates T_j, and that this structure is public. Section 3.2 concedes that defining these quantities is 'perhaps the most challenging part' of using the definition and that no empirical technique is provided. If secrets cannot be identified accurately in real corpora, the formal protection applies to the wrong objects and the practical value of the method collapses. A second fragile premise is the cited dominating pair from [CGM+24] used for noise calibration.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes a relaxation of differential privacy for the same-owner setting, where the goal is to bound the posterior probability that an adversary reconstructs each of a set of known secrets, with per-secret privacy budgets. It defines {p_j, r_j}-secret protection (Definition 3.1), proves a KL/f-divergence-based reconstruction bound (Lemma 3.2), and proposes an algorithm that assigns example weights via a linear program, then applies Poisson sampling with DP-SGD while calibrating the per-secret noise multiplier against a dominating pair. The empirical sections compare this pipeline with running DP-SGD on the full dataset, reporting roughly an 8x reduction in noise multiplier and about 7.5% test-loss improvement, and also present a separate study on noise and memorization in pretrained LLMs.

Significance. If the formal claims held, the definition and algorithm would be a useful contribution for settings where the model trainer also owns the data and wants to protect a small set of known secrets rather than all user-level information. Lemma 3.2 is a clean and apparently correct generalization of Fano-type reconstruction bounds, and the calibration is not circular: the KL target mu_j is fixed by the desired (p_j, r_j), while the LP constant c only shapes the sampling probabilities and does not affect the formal validity of the final guarantee. The paper is also honest about the hard open problem of specifying S, E, and T_j in Section 3.2. However, the central delivery is currently compromised by two accounting issues: the replacement-sensitivity mismatch between Definition 3.1 and the dominating pair used in Algorithm 2, and the unjustified 10x scaling assumption in Section 5.1. These issues affect whether the reported noise multipliers actually deliver the promised r_j bounds and whether the empirical claims are supported as written.

major comments (3)
  1. [§4, Algorithm 2 step 4; Definition 3.1] The adjacency in Definition 3.1 is replacement-based: an input differing only in secret j is obtained by replacing examples containing y_j with examples satisfying T_j(x_i, x_i')=1. Under per-example clipping to norm C, replacing one sampled example can change the batch sum by up to 2C, not C. The dominating pair used in Algorithm 2 step 4, N(\sum_{x_i \in D_j} Bern(rho_i), sigma^2)^T versus N(0, sigma^2)^T, is the standard add/remove pair for sensitivity C. Unless Theorem 1 of [CGM+24] is explicitly stated for replacement adjacency with the factor 2 absorbed into C or sigma, the binary search returns a sigma that is roughly 2x too small, and the claimed per-secret reconstruction bounds are not guaranteed. Please state the exact theorem used and, if replacement adjacency is intended, use the pair N(2\sum_{x_i \in D_j} Bern(rho_i), sigma^2)^T versus N(0, sigma^2)^T or otherwise account for the 2C sensitivity.
  2. [§5.1, footnote 1] The statement that scaling the batch size and dataset size by 10 is 'equivalent to reducing the noise multiplier by a factor of 10' is not justified and conflicts with standard DP-SGD accounting. In the Poisson-sampled Gaussian mechanism, the privacy loss per round depends on the sampling rate p = B/N and on the noise multiplier sigma, not on the absolute sizes of B and N; scaling both by 10 leaves p unchanged, and for a fixed number of rounds T the required sigma is unchanged. If the intention was to model the effect of a larger batch on gradient-estimate noise, that is a utility effect rather than a privacy-accounting effect and should be stated as such. As written, the noise multipliers in Figure 1c and the resulting 7.5% test-loss improvement are not supported. Please either run the experiments at the claimed scale or redo the accounting without this ad hoc reduction.
  3. [§3.2] Section 3.2 explicitly concedes that defining S, E, and T_j is 'perhaps the most challenging part' of using the definition and states that no empirical technique for doing so is provided. Since the experiments treat arbitrary words as secrets and set T_j identically to 1, the evaluation does not demonstrate that the framework can be applied to realistic secret structures where containment and transformation predicates must be inferred. This is a scope limitation rather than a mathematical error, but it is load-bearing for the paper's practical claim and should be stated as a prominent caveat in the abstract and conclusions rather than only in the body.
minor comments (5)
  1. [§4, after equation (1)] The text refers to the graph G as public, but G is not defined; it should be defined as the bipartite incidence graph of the containment function E.
  2. [Algorithm 2, step 4] The notation D_j is used without definition; please define D_j = E^{-1}(y_j) = {x in D : y_j in E(x)} explicitly in the algorithm or its preamble.
  3. [Lemma 3.2, proof] The proof is correct but compressed, especially the passage 'abusing notation to let a random variable denote a distribution'; a more explicit statement of the joint sampling process would improve readability.
  4. [§5.1, footnote 1] The sentence 'Since we decrease the scale of the noise, this should only make any comparisons more unfavorable for our LP-based approach' is unclear, because the preceding sentence claims an equivalence rather than an actual change; please clarify whether the sigma values in Figure 1c are the actual training values or simulated values.
  5. [§5.2] The pretraining memorization experiments are exploratory and are connected to Definition 3.1 only informally; please label this section as a separate empirical study rather than a validation of the proposed secret-protection definition.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the per-secret guarantee is calibrated from the target (p_j, r_j) and benchmarked against a fresh DP-SGD baseline, not derived from its own conclusion.

full rationale

The derivation chain is not circular. Definition 3.1 fixes target parameters (p_j, r_j), and Lemma 3.2 proves a sufficient divergence-based condition for the reconstruction bound. Algorithm 2 computes each µ_j = KL(Bern(r_j), Bern(p_j)) from the target and then chooses σ_j as the minimum noise such that the KL divergence of the dominating pair is at most µ_j; this is calibration, not a fitted prediction, and the claimed r_j bound holds by the lemma's proof for any σ satisfying the inequality. The LP constant c is tuned, but the guarantee is independent of c because σ is re-calibrated to the realized sampling probabilities. The empirical improvement is measured against DP-SGD on the full dataset, a separate algorithm, not against a re-statement of the paper's own result. The main external input is Theorem 1 of [CGM+24], a formal dominating-pair result from prior work with overlapping authorship; it is cited as a theorem and the KL quantities are computed with the open-source dp accounting library, so it does not reduce to the paper's own claim. The sensitivity-factor concern raised in the skeptic note is a potential correctness or accounting issue, not an instance of circular reasoning, because the paper explicitly defines the adjacency relation and claims a dominating pair for it; whether the pair truly dominates is a mathematical verification question, not a definitional equivalence.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claim does not introduce new physical or mathematical entities; 'secrets', predicates T_j, and containment function E are formal objects defined by the paper, not postulated unobserved entities. The load-bearing assumptions are the cited dominating pair, the public secret-structure assumption, and the scaling equivalence used in experiments.

free parameters (1)
  • LP constant c = tuned over c=2^k, k in {-6,...,4}; best value not stated
    Trades off example retention against noise multiplier; selected empirically in Section 5.1, so the reported utility gains are partly dependent on this choice.
assumptions (4)
  • standard math DP-SGD with Poisson sampling has a dominating pair N(sum_i Bern(rho_i), sigma^2)^T for inputs differing only in examples containing a given secret (Theorem 1 of [CGM+24]).
    Invoked in Section 4 to justify noise calibration; if this dominating pair does not hold for the multi-secret public-graph setting, the secret-protection guarantee is not established.
  • domain assumption DP-SGD's group privacy parameter is roughly linear in the group size, justifying the linear per-secret constraint in LP (1).
    Stated in Section 4; used to set mu_j-based caps, but the final sigma calibration uses actual sampling probabilities, so this assumption affects utility rather than the formal guarantee.
  • domain assumption The secret structure (S, E, T_j) is known and public; adjacent datasets share it.
    Assumed throughout Definition 3.1 and Section 4; Section 3.2 concedes that defining these objects is 'perhaps the most challenging part' and provides no automated method.
  • ad hoc to paper Scaling the batch size and dataset size by 10 is equivalent to reducing the noise multiplier by a factor of 10 in the privacy accounting.
    This equivalence in Section 5.1 is not proven and appears inconsistent with standard accounting, where the sampling rate q = B/n is unchanged when both are scaled.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hush! Protecting Secrets During Model Training: An Indistinguishability Approach." pith.science (2026). https://pith.science/paper/DVNHGSYI

@misc{pith2026250600201,
  author       = {Pith},
  title        = {Pith review of: Hush! Protecting Secrets During Model Training: An Indistinguishability Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DVNHGSYI}},
  note         = {Machine review of arXiv:2506.00201}
}
read the original abstract

We consider the problem of secret protection, in which a business or organization wishes to train a model on their own data, while attempting to not leak secrets potentially contained in that data via the model. The standard method for training models to avoid memorization of secret information is via differential privacy (DP). However, DP requires a large loss in utility or a large dataset to achieve its strict privacy definition, which may be unnecessary in our setting where the data curator and data owner are the same entity. We propose an alternate definition of secret protection that instead of targeting DP, instead targets a bound on the posterior probability of secret reconstruction. We then propose and empirically evaluate an algorithm for model training with this secret protection definition. Our algorithm solves a linear program to assign weights to examples based on the desired per-secret protections, and then performs Poisson sampling using these weights. We show our algorithm significantly outperforms the baseline of running DP-SGD on the whole dataset.

Figures

Figures reproduced from arXiv: 2506.00201 by the authors.

Figure 1
Figure 1. Comparison of LP solutions for different values of [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Effect of noise on the validation loss 0 0.001 0.005 0.01 0.02 Noise std σ 0.005 0.010 0.015 0.020 Fraction Extraction Gemma2 250M Gemma2 20M Gemma2 2B [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

24 extracted references · 16 canonical work pages

  1. [1]

    Goodfellow, H

    Mart \' n Abadi, Andy Chu, Ian J. Goodfellow, H. Brendan McMahan, Ilya Mironov, Kunal Talwar, and Li Zhang. Deep learning with differential privacy. In Proc. of the 2016 ACM SIGSAC Conf. on Computer and Communications Security ( CCS '16) , pages 308--318, 2016

  2. [2]

    Reconstructing training data with informed adversaries

    Borja Balle, Giovanni Cherubin, and Jamie Hayes. Reconstructing training data with informed adversaries. pages 1138--1156, 05 2022

  3. [3]

    Protection against reconstruction and its applications in private federated learning, 2019

    Abhishek Bhowmick, John Duchi, Julien Freudiger, Gaurav Kapoor, and Ryan Rogers. Protection against reconstruction and its applications in private federated learning, 2019

  4. [4]

    Private empirical risk minimization: Efficient algorithms and tight error bounds

    Raef Bassily, Adam Smith, and Abhradeep Thakurta. Private empirical risk minimization: Efficient algorithms and tight error bounds. In Proc. of the 2014 IEEE 55th Annual Symp. on Foundations of Computer Science (FOCS) , pages 464--473, 2014

  5. [5]

    Clement, Matthew Bierbaum, Kevin P

    Colin B. Clement, Matthew Bierbaum, Kevin P. O'Keeffe, and Alexander A. Alemi. On the use of arxiv as a dataset. CoRR , abs/1905.00075, 2019

  6. [6]

    Rachel Cummings and Deven R. Desai. The role of differential privacy in gdpr compliance. 2018

  7. [7]

    Mind the privacy unit! user-level differential privacy for language model fine-tuning

    Lynn Chua, Badih Ghazi, Yangsibo Huang, Pritish Kamath, Ravi Kumar, Daogao Liu, Pasin Manurangsi, Amer Sinha, and Chiyuan Zhang. Mind the privacy unit! user-level differential privacy for language model fine-tuning. In First Conference on Language Modeling , 2024

  8. [8]

    Brendan McMahan, Nicole Mitchell, Krishna Pillutla, and Keith Rush

    Zachary Charles, Arun Ganesh, Ryan McKenna, H. Brendan McMahan, Nicole Mitchell, Krishna Pillutla, and Keith Rush. Fine-tuning large language models with user-level differential privacy, 2024

Show all 24 references
  1. [9]

    Quantifying memorization across neural language models

    Nicholas Carlini, Daphne Ippolito, Matthew Jagielski, Katherine Lee, Florian Tramer, and Chiyuan Zhang. Quantifying memorization across neural language models. In The Eleventh International Conference on Learning Representations

  2. [10]

    Extracting training data from large language models

    Nicholas Carlini, Florian Tram \`e r, Eric Wallace, Matthew Jagielski, Ariel Herbert-Voss, Katherine Lee, Adam Roberts, Tom Brown, Dawn Song, \'U lfar Erlingsson, Alina Oprea, and Colin Raffel. Extracting training data from large language models. In 30th USENIX Security Sympos...

  3. [11]

    BERT : Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT : Pre-training of deep bidirectional transformers for language understanding. In Jill Burstein, Christy Doran, and Thamar Solorio, editors, Proceedings of the 2019 Conference of the North A merican Chapter ...

  4. [12]

    R2t: Instance-optimal truncation for differentially private query evaluation with foreign keys

    Wei Dong, Juanru Fang, Ke Yi, Yuchao Tao, and Ashwin Machanavajjhala. R2t: Instance-optimal truncation for differentially private query evaluation with foreign keys. In Proceedings of the 2022 International Conference on Management of Data , SIGMOD '22, page 759–772, New York,...

  5. [13]

    Calibrating noise to sensitivity in private data analysis

    Cynthia Dwork, Frank McSherry, Kobbi Nissim, and Adam Smith. Calibrating noise to sensitivity in private data analysis. In Proc. of the Third Conf. on Theory of Cryptography (TCC) , pages 265--284, 2006

  6. [14]

    Google's differential privacy libraries., 2022

    DP Team . Google's differential privacy libraries., 2022. https://github.com/google/differential-privacy

  7. [15]

    It's my data too: Private ml for datasets with multi-user training examples, 2025

    Arun Ganesh, Ryan McKenna, Brendan McMahan, Adam Smith, and Fan Wu. It's my data too: Private ml for datasets with multi-user training examples, 2025

  8. [16]

    Bounding training data reconstruction in DP - SGD

    Jamie Hayes, Borja Balle, and Saeed Mahloujifar. Bounding training data reconstruction in DP - SGD . In Thirty-seventh Conference on Neural Information Processing Systems , 2023

  9. [17]

    Preventing verbatim memorization in language models gives a false sense of privacy

    Daphne Ippolito, Florian Tram \`e r, Milad Nasr, Chiyuan Zhang, Matthew Jagielski, Katherine Lee, Christopher A Choquette-Choo, and Nicholas Carlini. Preventing verbatim memorization in language models gives a false sense of privacy. arXiv preprint arXiv:2210.17546 , 2022

  10. [18]

    Conservative or liberal? personalized differential privacy

    Zach Jorgensen, Ting Yu, and Graham Cormode. Conservative or liberal? personalized differential privacy. volume 2015, 04 2015

  11. [19]

    Scalable extraction of training data from (production) language models

    Milad Nasr, Nicholas Carlini, Jonathan Hayase, Matthew Jagielski, A Feder Cooper, Daphne Ippolito, Christopher A Choquette-Choo, Eric Wallace, Florian Tram \`e r, and Katherine Lee. Scalable extraction of training data from (production) language models. arXiv preprint arXiv:23...

  12. [20]

    The fineweb datasets: Decanting the web for the finest text data at scale, 2024

    Guilherme Penedo, Hynek Kydlíček, Loubna Ben allal, Anton Lozhkov, Margaret Mitchell, Colin Raffel, Leandro Von Werra, and Thomas Wolf. The fineweb datasets: Decanting the web for the finest text data at scale, 2024

  13. [21]

    Stochastic gradient descent with differentially private updates

    Shuang Song, Kamalika Chaudhuri, and Anand D Sarwate. Stochastic gradient descent with differentially private updates. In 2013 IEEE Global Conference on Signal and Information Processing , pages 245--248. IEEE, 2013

  14. [22]

    Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, Johan Ferret, Peter Liu, Pouya Tafti, Abe Friesen, Michelle Casbon, Sabela Ramos, Ravin Kumar, Charline Le Lan...

  15. [23]

    Individual privacy accounting for differentially private stochastic gradient descent

    Da Yu, Gautam Kamath, Janardhan Kulkarni, Tie-Yan Liu, Jian Yin, and Huishuai Zhang. Individual privacy accounting for differentially private stochastic gradient descent. Transactions on Machine Learning Research , 2023

  16. [24]

    Optimal accounting of differential privacy via characteristic function

    Yuqing Zhu, Jinshuo Dong, and Yu-Xiang Wang. Optimal accounting of differential privacy via characteristic function. In Gustau Camps-Valls, Francisco J. R. Ruiz, and Isabel Valera, editors, Proceedings of The 25th International Conference on Artificial Intelligence and Statist...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.