- Rating
- 3
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper proposes a hierarchical diffusion framework for function generation. A linear encoder decomposes each function as , where represents global structure and is the residual. A finite-dimensional diffusion model first generates the global state, followed by a conditional residual diffusion model that uses the noisy query value and fixed-dimensional local polynomial statistics. The theoretical analysis decomposes the residual denoising error relative to the Bayes predictor into discretization, localization, compression, and learning errors, and derives conditional and end-to-end Wasserstein bounds under stated assumptions. Evaluation includes a hierarchical Gaussian process mixture benchmark with analytically available Bayes denoisers, CIFAR-10 image generation, and signed distance fields derived from MNIST and Fashion-MNIST. These experiments examine information loss from localization and compression, the influence of global-stage errors, and the accuracy and model-size trade-offs between local and full-context residual denoisers. The image experiments also consider variants with learned neighborhood features.
Critical strengths
-
A useful framework for understanding information loss in functional denoising. The hierarchical formulation distinguishes uncertainty about global structure from residual dependence after conditioning on the global state. Theorem 1 then gives an exact decomposition into discretization, localization, compression, and learning errors, without requiring conditional Gaussianity for this identity. This is consequential because it separates limitations of the available information from limitations of the learned predictor, providing an interpretable framework for evaluating local representations (Section 4.1).
-
Controlled experiments that directly examine the proposed mechanism. The Gaussian process mixture benchmark provides analytically available Bayes denoisers under different information sets, allowing localization and compression losses to be measured separately from neural-network learning error. Its nonstationary and anisotropic settings make this a useful diagnostic of when local compression works (Section 5.1; Figures 1–2). The signed-distance-field experiments provide complementary practical evidence: on the datasets, the local residual model uses 0.12M parameters versus 2.5M for the full-context baseline and achieves lower held-out denoising errors at the reported noise levels (Table 5). Together, these results support the usefulness of compact local conditioning.
Critical weaknesses
-
The end-to-end bound has materially different discretization dependence in the main text and the proof. Theorem 10 gives localization and compression terms of and , whereas Equation (39) in Appendix D.10 multiplies these by and , respectively. Although the main text refers to the appendix for the precise form, these factors cannot be absorbed into constants when studying refinement in the context size . This discrepancy is consequential because the end-to-end guarantee is a central contribution, and the omitted dependence changes what the bound establishes as the discretization is refined. The main statement should explicitly retain this dependence or specify that the design is fixed.
-
The image experiments do not establish the claimed reduction in global-state sensitivity. Section 5.2, finding (i), interprets the reduction of FID(gen) minus FID(oracle) from 27.4 to 18.5 as a smaller . However, Theorem 4 is an upper bound involving function-space Wasserstein distances; the difference between two FID values does not identify this conditional Lipschitz constant. Table 2 illustrates the ambiguity: adding noise at sampling reduces the gap from 18.5 to 11.6 while worsening gen/oracle FID from 30.8/12.3 to 48.8/37.1, as the authors acknowledge. Thus, the reported diagnostic does not substantiate the claimed stability mechanism. This limits the strength of the theory-to-experiment connection, which is the stated purpose of the image experiments.
Questions for authors
-
Is Theorem 10 intended only for a fixed context design, or does it describe how the error behaves as the context size increases? Please reconcile its localization and compression terms with the additional factors and in Equation (39). Which quantities are held fixed, and what conclusion about discretization refinement follows from the bound with all dependencies retained? This would clarify the scope of the main theoretical guarantee.
-
What justifies the inference in Section 5.2(i) that a smaller FID(gen) minus FID(oracle) implies a smaller ? In particular, how does this inference account for Table 2, where sampling-time perturbations reduce the gap while worsening both FIDs? Is there evidence directly measuring the conditional sampler's sensitivity to changes in the global state, beyond the FID gap? This would help determine whether the experiments validate the proposed stability mechanism or establish an improvement in sample quality without identifying that mechanism.
-
In Appendix B.3, do the and results use the same trained residual-network weights, or separately trained models with the same parameter count? Please clarify whether the demonstrated resolution independence concerns architecture size, transfer of a trained sampler across grids, or both. If the same weights are reused, how are the global coefficients, local window scales, and residual noise defined consistently across resolutions?
4 comments
W1 / Q1 (Theorem 10 and its dependence on ): The main text Theorem 10 omits the factors and of Equation (39). Is it intended only for a fixed design, which quantities are held fixed and what follows under refinement?
Response: Theorem 10 is a bound for a fixed design . Following your suggestion, the current revision specifies this in the statement ("a fixed quasi-uniform design "), and the paragraph after it (in blue) states the two factors of (39) and explains where they come from. The new Appendix E adds a bound that does not grow with . Both bounds are meaningful, since Theorem 10 does not need the kernel condition of Appendix E.
-
What is held fixed for Theorem 10. The design , the radius , the degree and the stopping time are fixed (Appendix D.10). The theorem describes the error as the number of training functions grows: the three statistical terms tend to zero, and what remains is the approximation error of this design and the early-stopping error.
-
Reconciling Theorem 10 with (39). In (39) the localization term is and the compression term is . In Theorem 10 the two factors are constants for a fixed design, so the main statement shows only the dependence on . As you note, they cannot be absorbed when varies, so the text after Theorem 10 now states them. Both errors are energy norms of kriging weights: for localization (Theorem 5) and for compression (Proposition 8). The proofs of Theorem 6 and Proposition 9 bound them through the Euclidean norm , which brings in and .
-
What follows under refinement. With all dependencies retained, (39) grows with , so it gives no rate as the design is refined. This growth is in the bound, not in the error.
- The approximation error is bounded uniformly in the design. By Theorem 1, the discretization, localization and compression errors sum to for every design. So the three approximation terms of (39) can be replaced by , and the bound stays bounded as (new Proposition 14, Appendix D.10).
- Appendix E gives a bound that does not grow with . Under refinement grows, while the energy norm of the kriging weights and compression error saturate (Figure 6). The energy norm is a variance reduction, bounded uniformly in the design (Lemma 2), and Appendix E works in that norm. It assumes that the Bayes denoiser is an integral of against a square-integrable kernel , which holds for stationary Matérn fields with suitable smoothness orders (Proposition 15) and for our finite-rank benchmark (Section 5.1). The three approximation terms then become (Proposition 17): a quadrature error that vanishes under refinement, the tail of outside the window and a compression error without the factor . So at fixed and the bound converges as to a floor set by the window and the degree. This floor is the price of computational efficiency: the denoiser sees only a -dimensional summary of an infinite-dimensional , so the input dimension of the network and its cost per query do not grow with , and a finite summary loses some information in general. In Figure 1 each approximation term is at most a few percent of the residual variance.
- The measured error does not grow either. Without retraining, the denoising error for stays within 2.4% from to (Table 5).
Together with our first comment, this responds to every weakness and question in your review, and we hope it addresses your concerns. We thank you again for your careful reading and valuable suggestions.
Thank you for the clarification. I will maintain my rating of 3 (weak acceptance).
Thank you for your acknowledgement and for the time you spent on our paper. We appreciate your support and constructive suggestions.
We sincerely thank Reviewer LJKm for the thoughtful review and the precise questions on the dependence of Theorem 10 on m and on what the FID gap can identify. We are glad that the reviewer finds the information-loss framework useful for understanding functional denoising, and values the controlled experiments that examine the proposed mechanism directly.
Following your suggestions, we now measure the sensitivity Lgen(θ) directly and test resolution transfer with a single trained model, marked in blue in the revised PDF.
W2 / Q2 (FID gap and Lgen(θ)): A smaller FID gap does not identify a smaller Lgen(θ), and Table 2 shows sampling noise lowering the gap while worsening both FIDs.
Response: We agree. Theorem 4 bounds the Wasserstein distance between the generated and true function distributions, and a difference of FIDs cannot identify a conditional Lipschitz constant. Following the reviewer's suggestion, we now estimate the sampler's sensitivity Lgen(θ) to the global state directly with a coupling probe L^ (Section 5.2, Appendix B.2, Tables 2 and 4).
Q3 (same weights for 282 and 642?): It is unclear whether the 282 and 642 SDF results use the same weights, and how the global coefficients, window and noise are defined across resolutions.
Response: In the submitted version they did not. The 642 model was trained separately with the same parameter count. The revised experiment now uses one set of weights, trained on 282 only and used on every grid without retraining. The updated SDF results are in Section 5.3, Appendix B.3, Table 5 and Figures 13–14.
We address W1 / Q1 (Theorem 10 and its dependence on m) in a separate comment.