- Rating
- 2
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper proposes DP-DeepSets, a hypernetwork trained on public data to generate model parameters from private datasets. Instead of repeatedly adding noise to gradients as in DP-SGD, the method adds noise once to an aggregated dataset embedding and uses it to generate model parameters.
Critical strengths
Using a public hypernetwork to generate private model parameters is a promising way to avoid repeated noise injection during DP training. The theoretical analysis also provides some insight into why learning low-dimensional task representations could help.
Critical weaknesses
1. The experimental evaluation is very limited.
I do not understand why diffusion-model fine-tuning was chosen as the only evaluation task. The paper proposes a general DP learning method, but evaluates only one task and one target-model architecture. Why not test on standard DP benchmarks such as CIFAR-10 classification? This would make it easier to compare against existing DP methods and understand the actual improvement.
2. The baselines are not sufficiently strong.
The paper mainly compares against relatively basic DP optimization methods. Stronger DP-SGD training recipes, such as those in Unlocking High-Accuracy Differentially Private Image Classification through Scale (De et al., 2022, https://arxiv.org/abs/2204.13650), should be considered on a suitable classification benchmark.
I am also concerned about the baseline hyperparameters in Appendix E. For example, Table 2 reports unusually large learning rates for DP-SGD. The authors should clarify these settings and demonstrate that the baselines are properly tuned.
3. Important reference results are missing.
The paper should report the performance of:
a. The original pretrained model without fine-tuning.
b. Non-private fine-tuning on the same private datasets.
c. Fine-tuning using only public data, without private adaptation.
Without these results, it is difficult to know how good the proposed method actually is, how much utility is lost due to privacy, and how much improvement comes from private data rather than public training.
4. The privacy budgets are questionable.
The experiments include privacy budgets as large as . Such values provide little meaningful privacy protection. More concerningly, DP-SGD still performs poorly at these extremely large budgets. This raises questions about whether the baseline is properly optimized. The evaluation should focus more on meaningful privacy budgets and include non-private reference results.
5. The scope of the privacy guarantee is unclear.
The privacy proof considers a fixed input dataset, while the experiments construct overlapping private datasets using k-NN and generate multiple models. It is unclear whether the reported privacy guarantee applies to individual 128-example tasks or to the entire data-processing and model-release procedure.
Questions for authors
-
Why diffusion models? Can the authors evaluate the method on standard classification tasks such as CIFAR-10, preferably with more than one model architecture?
-
Can the authors provide stronger baselines? In particular, how do the results compare with carefully tuned DP-SGD methods? Are the unusually large learning rates in Appendix E correct?
-
Can the authors report non-private and public-only results? These are essential for understanding the actual privacy-utility trade-off. How much does conditioning on the private dataset improve over using public data alone?
-
Why evaluate at ? Is this intended to approximate non-private training? If so, why does DP-SGD still perform poorly?
-
What is the exact privacy guarantee? Does it cover the entire experimental pipeline, including k-NN task construction and multiple model releases, or only each fixed-size private task?
6 comments
Response to Reviewer Ftcg (2/2)
W3 & Q3: Reference results
a. The original pretrained model without fine-tuning. b. Non-private fine-tuning on the same private datasets. c. Fine-tuning using only public data, without private adaptation.
We have added all three, evaluated on the same 100 private tasks with 1,024 generated images each. (a) is the ImageNet64 checkpoint without LoRA. (b) trains the same rank-1 LoRA with AdamW on each private task without clipping or noise, with its learning rate and number of updates selected on public tasks. (c) is the model fine-tuned on all public examples, which the Public FT + DP-SGD baseline starts from, without private adaptation. FID (mean SD):
| Model | FID |
|---|---|
| (a) Pretrained model | 251.4 19.6 |
| (c) Public-only fine-tuning | 230.4 29.1 |
| Ours, | 137.4 24.3 |
| Ours, | 133.4 23.2 |
| (b) Non-private fine-tuning | 106.6 21.7 |
Public-only fine-tuning lowers the FID of the pretrained model by only 21, whereas conditioning on the private dataset lowers it by a further 93 at , which closes 75% of the gap to non-private fine-tuning. At , the FID of our method is 30.8 higher than that of non-private fine-tuning, whereas that of DP-AdamW, which uses the same optimizer, is 144.5 higher (FID 251.1). Section 5 also reports non-private results for classification.
W4 & Q4: Privacy budgets and
The experiments include privacy budgets as large as . Such values provide little meaningful privacy protection. More concerningly, DP-SGD still performs poorly at these extremely large budgets.
Is this intended to approximate non-private training? If so, why does DP-SGD still perform poorly?
Figure 2 sweeps from 0.125 to 1024 to show how each method behaves as the privacy constraint is relaxed, and is the end of this sweep. It is not meant to approximate non-private training, and our main claims concern , as in Table 1 () and the new classification experiments ( from 0.25 to 8). As the non-private reference, we have added fine-tuning without clipping or noise (W3).
We agree that the submitted DP-SGD was not tuned well enough. In particular, its was too large for gradients of median norm about . Re-tuning on public tasks reduced to and the FID at from 215.3 to 152.1, close to DP-AdamW with a tuned (153.5). Even at , the 800 updates on only 128 examples require a noise multiplier of 0.69. Since the noise is added to all 40,704 LoRA parameters, its norm in each averaged update is about , larger than that of the averaged clipped gradient (at most ). Our method at every from 1 to 128 (FID between 132.8 and 137.4) is better than DP-SGD at .
W5 & Q5: Scope of the privacy guarantee
It is unclear whether the reported privacy guarantee applies to individual 128-example tasks or to the entire data-processing and model-release procedure.
The guarantee applies to each 128-example task. Our mechanism takes one fixed task as input and outputs one model, and its -DP guarantee protects each example of that task. It does not cover the -NN construction and the release of all 100 models jointly. The -NN construction is only our way of creating many small private datasets for evaluation, as in a setting where each data owner holds one small dataset, and all baselines are given the same per-task guarantee. If the private pool were instead regarded as a single dataset, an image used in several tasks would incur composition over those releases, and the construction itself would also have to be private. We do not claim such a guarantee, and we have stated the scope explicitly in Section 5.
We would be happy to clarify any remaining concerns during the discussion period. We would greatly appreciate it if you could consider increasing the score once all concerns have been resolved.
Further revised PDF and summary of revisions
Thank you again for your review. We have uploaded a further revised PDF in which DP-SGD in Figure 2 is re-tuned at all budgets. Our method remains better at every budget from to , where both are evaluated, e.g., FID 132.8 versus 251.9 at . At large budgets, the re-tuned DP-SGD is now close to DP-AdamW, e.g., 152.1 versus 153.5 at .
The main revisions in response to your review are as follows, with details in our response above.
- Classification (W1 & Q1). Section 5 now includes CIFAR-10 and CIFAR-100, where our method outperforms DP-SGD at every budget from to , e.g., 98.4% versus 78.5% on CIFAR-10 (binary) at .
- Re-tuned DP-SGD (W2 & Q2, W4 & Q4). In both tasks, the learning rate, the number of updates, the clipping threshold, and the batch size are selected on public tasks (Appendices E and G). The candidate batch sizes are 32 and 128, the full private dataset, since large batches often help DP-SGD (De et al., 2022). Appendix E also explains why the selected learning rates of and are large.
- Reference results (W3 & Q3). The pretrained model, public-only fine-tuning, and non-private fine-tuning attain FID 251.4, 230.4, and 106.6, compared with 137.4 for our method at (Appendix E).
- Scope of the privacy guarantee (W5 & Q5). Section 5 now states that the guarantee applies to each 128-example task, not to the -NN construction or the joint release of all 100 models.
We would be grateful to know whether these revisions address your concerns, and we would be happy to clarify anything further before the discussion period ends. If they do, we would appreciate it if you could consider updating your score.
Response to Reviewer Ftcg (1/2)
We thank the reviewer for the careful reading and the concrete suggestions. We address the concerns and questions below; all revisions mentioned here are included in the revised PDF.
W1 & Q1: Classification experiments and the choice of diffusion models
New classification experiments. Following the suggestion, we have added classification on CIFAR-10 (binary) and CIFAR-100 (10-way) to Section 5. Since our target is adapting a pretrained model to a small private dataset, each private task again has 128 examples, and private tasks use class pairs (CIFAR-10) or classes (CIFAR-100) unseen during public training. This differs from the standard benchmark that uses the full CIFAR-10 training set, so the accuracies are not comparable to those reported there. With only 128 examples, DP-SGD needs far more noise relative to the gradient signal than in the full-dataset setting. At ϵ=1 and δ=10−5, accuracy over 100 private tasks is 98.4% versus 78.5% for DP-SGD on CIFAR-10, and 34.3% versus 12.3% on CIFAR-100. Our method is better at all six budgets from ϵ=0.25 to 8. The full table is in Section 5 and in our response to Reviewer n6F4.
More than one architecture. The hypernetwork generates a linear classifier on frozen CLIP features, so the target models now include both the LoRA of a diffusion U-Net and a linear classifier.
Why diffusion models. We chose diffusion-model fine-tuning because generating low-rank weights of a text-to-image diffusion model from reference images is a known use of hypernetworks (Ruiz et al., 2024).
W2 & Q2: Stronger baselines and the learning rates
Stronger and re-tuned DP-SGD. In both tasks, we have re-tuned DP-SGD on public tasks only, including large batches up to the full 128 examples as recommended by De et al. (2022), together with the learning rate, the number of updates, and the clipping threshold. In classification, where DP-SGD trains the same linear head on the same frozen features, the gap remains large, e.g., 52.8% versus 15.8% on CIFAR-100 at ϵ=2. In generation, the re-tuned FIDs at ϵ=1 and 4 are 369.0 and 257.0, while ours are 137.4 and 133.4, so our method remains better by more than 120 in FID. Appendix E gives the details.
Learning rates of 102 and 103. Yes, the values are correct. Each DP-SGD update has norm at most the learning rate times the clipping threshold C, apart from noise, and the per-example gradients of rank-1 LoRA are small (median norm about 1.7×10−3 on public tasks). A learning rate of 103 therefore gives updates of moderate size. DP-AdamW, which normalizes the gradient scale, selects ordinary rates of 0.003–0.01.