- Rating
- 1(−1)
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper studies linear Flow Matching when the target observations follow the noisy latent-generator model. Although the target distribution has full -dimensional support for every fixed , the authors show that the nonlinear part of the population FM velocity can be represented through a posterior mean over the -dimensional latent variable. They construct a spatially regular ReLU approximation whose leading resolution complexity scales as , prove a high-probability integrated velocity-risk bound with leading rate , and propagate this error through the learned ODE to obtain a convergence rate with leading factor . The paper also gives an Euler-discretization bound under an additional temporal-regularity assumption.
The result should not be interpreted as learning to remove the target noise or recover the clean law of . The target law in the analysis is the noisy, full-dimensional distribution itself. The contribution is instead an intrinsic-dimensional statistical analysis of FM without requiring exact low-dimensional support or a separately learned latent representation.
Critical strengths
-
A technically interesting extension beyond exact low-dimensional support. The paper proves that the leading approximation and estimation exponents can depend on the latent dimension even though the observed distribution is full-dimensional. This addresses a gap between exact-manifold theory and noisy latent-variable models and yields the rate under the paper's assumptions.
-
The approximation class is designed for both estimation and transport. The authors do more than approximate the population velocity in value: they carry a spatial Lipschitz profile through the same neural construction and use it to convert integrated velocity error into a non-asymptotic guarantee. This makes the statistical result relevant to the generated distribution rather than stopping at a regression bound.
Critical weaknesses
-
The paper does not establish that the latent-dimensional exponent yields meaningful sample efficiency in a practically relevant regime. Theorem 4.5 chooses but also requires conditions (5)--(6). For example, already implies , while the other conditions impose additional restrictions involving , , and , including terms that grow polynomially with and deteriorate as decreases. Moreover, the network width and sparsity scale polynomially with . Thus the result has a latent-dimensional exponent but not a uniformly latent-dimensional sample or computational complexity. The paper neither translates these assumptions into a clear non-vacuous high-dimensional regime nor demonstrates that the favorable exponent dominates the ambient-dependent thresholds and constants at realistic sample sizes. This substantially weakens the significance of the central claim.
-
The finite-sample theorem does not analyze the FM objective used in implementable training. In equation (9), the expectation over the Gaussian source and the integral over time are evaluated exactly inside each per-target loss. Consequently, the finite-sample randomness comes only from the target observations. Practical FM instead samples and and optimizes the resulting stochastic objective. The theorem therefore omits auxiliary Monte Carlo error, optimization error, finite minibatches, and the difficulty of finding a global ERM under the imposed Lipschitz constraint. Although the paper acknowledges finite auxiliary sampling as future work, this gap makes it difficult to interpret the result as a sample-complexity guarantee for a trainable FM procedure.
-
The paper gives no evidence that its asymptotic phenomenon is observable, even in a controlled setting. I do not view experiments as mandatory for every theoretical paper. Here, however, the main significance question is precisely whether the favorable dependence on survives the large - and -dependent constants and applicability thresholds. A small synthetic experiment varying could test whether the predicted slopes appear, when ambient dependence dominates, and how the exact-integrated objective differs from sampled training. Without either such evidence or a sharper theoretical regime analysis, the practical meaning of the main rate remains unresolved.
-
The Euler result depends on a temporal-regularity condition that is not derived from the learned class. Assumption 5.3 requires a deterministic , independent of , controlling the time variation of the learned ERM. Section 3 explicitly imposes no temporal-regularity constraint on the class. The Euler proposition is valid as a conditional statement, but it is unclear whether the proposed network class and ERM satisfy this assumption with the claimed uniformity. The authors should either construct a class that enforces this property and track its complexity, or clearly present Proposition 5.4 as an additional conditional guarantee whose assumption is not established by the preceding learning theory.
Questions for authors
- Can the authors state a simple corollary giving sufficient growth conditions on and under which (5)--(6) hold and the advertised rate is non-vacuous?
- What additional term should one expect if the time variable and Gaussian source are approximated using a finite number of Monte Carlo samples per target observation?
- Does the constructed neural class imply Assumption 5.3 with a constant independent of ? If not, how should the Euler result be interpreted for the ERM analyzed in Theorem 4.5?
- Which parts of the posterior-network approximation are new relative to the noisy-latent score-matching machinery of Yakovlev and Puchkin (2025), beyond the clearly FM-specific velocity representation, excess-risk analysis, and ODE propagation?
5 comments
Response to Reviewer rBbA — 2. Finite auxiliary Monte Carlo samples
Thank you for pointing out the role of auxiliary sampling in an implementable objective. In Appendix F we now analyze a finite-sampling ERM over the same spatially constrained class as Theorem 4.5. The analysis follows the shared-auxiliary scheme of Zhou and Liu (2025): draw target observations and independent time–Gaussian pairs, then reuse those pairs for each target. Theorem 4.5 itself continues to concern the exact-average objective.
Let and denote the exact-average and sampled objectives, and set . Proposition F.1 controls the uniform difference of their anchored versions:
where the leading auxiliary term is
Here is the log covering number of the posterior-network class on the localized domain; Appendix F gives , , and the tail terms explicitly. The in the denominator reflects the independent shared pairs: the cross-product evaluations reuse both targets and auxiliary variables and are not independent target observations. Subtracting cancels the common unbounded affine contribution and does not change the minimizer.
Combining this deviation with the original target-sample concentration gives, with high probability,
where is the original target-sample remainder and is the integrated squared velocity error. Thus the auxiliary error is explicit and separate from the error due to the target observations. For fixed model parameters, set , , , and . The original comparator and covering bounds then yield , under the stated scale conditions. This is a sufficient budget for this particular shared-sampling bound.
In Appendix H, fresh time and Gaussian source variables are drawn at each U-Net minibatch update. Appendix F instead analyzes exact ERM for a fixed shared set of auxiliary pairs; these two settings provide complementary practical and statistical perspectives.
Zhou and Liu use the same decomposition, Gaussian truncation, and covering strategy for bounded-support targets and a different velocity class. The anchoring and target-tail localization in Appendix F adapt the argument to the present noisy-latent model.
Reference
- Zhou and Liu, 2025, An Error Analysis of Flow Matching for Deep Generative Modeling, Section 4.1 and Theorem 4.4.
Response to Reviewer rBbA — 3&4. Euler consistency and posterior-network novelty
Thank you for asking whether the discretization uses the same estimator as the learning theorem and which approximation ingredients are new. These questions clarify two distinct interfaces.
Changes in the revised manuscript: Section 3, Proposition 3.2, Appendix B.3–B.4, Section 5.2, Proposition 5.5, and Appendix G.1–G.2.
The estimator and Euler. Theorem 4.5 studies ERM in a class with a spatial envelope. The class does not itself establish Assumption 5.3 with a constant independent of . Proposition 5.4 gives a deterministic Euler guarantee when the estimator also satisfies this additional temporal condition. Proposition 5.5 connects the same estimator directly to a finite-step sampler using its existing spatial envelope. After training, draw independent and set , where . Writing and , its conditional output law, averaged over the numerical times and independent initialization, satisfies
The key step is . Spatial regularity controls the learned field along the numerical trajectory, and the required time regularity of the true field is derived from the noisy-latent model. Thus the result closes the discretization interface without adding temporal regularity to the learning class. The shared finite-auxiliary ERM in Appendix F can be used analogously after conditioning on its training randomness, with its risk controlled by the bound there.
Which posterior-network ingredients are shared or added? We use Yakovlev and Puchkin’s latent-cell Taylor and Gaussian-posterior machinery for the same noisy-generator model. The Gaussian posterior covariance identity and general first-order ReLU approximation techniques are also known tools. Our added construction is a single comparator whose posterior-value approximation and global spatial Lipschitz envelope hold together, with the leading latent-cell count retained and the full ambient/noise factors tracked. Its value approximation enters the FM risk; its spatial envelope places this same comparator in the class used for both constrained ERM and transport. Proposition 3.2 separately supplies the full-time generator-to-FM-velocity stability estimate, which connects the latent surrogate to that comparator. We have made these dependencies and the imported primitives explicit in Section 3 and Appendix B.
Related FM work by Fukumizu et al. also uses a spatially Lipschitz-constrained network class, for Besov-density targets and different interpolation assumptions. Our construction addresses the noisy latent-generator setting above.
The population score–velocity identity makes the relation to score matching transparent. Under the OU time change, full-time FM risk carries the weight , while the stated YP score-risk theorem is unweighted on a truncated interval. Additional estimates are consequently needed for a transfer of the full-time risk and the spatially constrained comparator.
References
Response to Reviewer rBbA — 1. Dimension, sample-size regimes, and the controlled experiment
Thank you for asking how the latent-dimensional rate behaves alongside the full dimension-dependent factors and scale conditions. We have added explicit sufficient growth regimes and a controlled image experiment that examines velocity learning as the ambient dimension increases.
Changes in the revised manuscript: Corollaries E.1–E.2 and Appendix H, Table 1; the original conditions are (5)–(6).
When are the conditions non-vacuous? Let s=2β+d, α=2β/s, and c=carch as in Theorem 4.5. Fix d,β,H, a positive σ, and δn=n−p. Corollary E.1 verifies every part of (5)–(6) eventually if, for example,
D_n\le Cn^\kappa,\qquad 0<\kappa<\min\left\\{\frac{2}{3s},\frac{\alpha}{c+5/2}\right\\}.
The corresponding integrated risk is at most Cn−α+(c+5/2)κ(logn)c+1 and the continuous-flow W2 bound tends to zero. In particular, Dn=O(logn) also satisfies the conditions. For a joint example, Corollary E.2 takes
Dn=⌈(logn)1/2⌉,σn=(logn)−3/8.
It gives noisy-target W2≤n−β/s+o(1) and an additional coupling bias O((logn)−1/8) to the clean generator law. These two corollaries give concrete sufficient regimes for continuous-flow consistency. Numerical sampling is described separately by Proposition 5.5, with explicit noise-dependent constants.
What do we observe in a controlled example? Appendix H uses a two-dimensional latent Gaussian-spot image family, σ=0.1, and a time-conditioned U-Net. Each entry is mean ± sample standard deviation over three training seeds; evaluation uses 16,384 held-out tuples per seed and a numerical posterior.
At fixed n=8192, the 3000-update risk changes from 0.01629 to 0.01761 as D increases sixteenfold, while the corresponding zero-posterior risks are 4.564, 4.420, and 4.488. With clean-signal norm fixed at 0.9, the expected squared noise norm increases from 2.56 to 40.96 across this comparison. The observed velocity risk thus remains stable across these resolutions in this noisy-latent image family. For d=2 and β=1, the first theoretical scale condition is n>D2; the experiment provides a complementary finite-budget view at smaller sample sizes for this structured instance.
At D=1024, the 3000-update risks are 0.01617, 0.01582, and 0.01608 as n increases from 8192 to 131072. The fixed-budget sample-size sweep is therefore approximately flat, while increasing the budget from 1000 to 3000 updates consistently lowers risk in all configurations. This experiment uses a time-conditioned U-Net with convolutional weight sharing and finite-step AdamW training; our theorems analyze exact ERM in a spatially constrained ReLU class. The fixed values of d and σ make the table a focused comparison of the tested resolutions and training budgets, complementing the dimension- and noise-dependent regimes in Appendix E.
Reference