Conditional score-based diffusion models for Bayesian inference in infinite dimensions

Since their initial introduction, score-based diffusion models (SDMs) have been successfully applied to solve a variety of linear inverse problems in finite-dimensional vector spaces due to their ability to efficiently approximate the posterior distribution. However, using SDMs for inverse problems in infinite-dimensional function spaces has only been addressed recently, primarily through methods that learn the unconditional score. While this approach is advantageous for some inverse problems, it is mostly heuristic and involves numerous computationally costly forward operator evaluations during posterior sampling. To address these limitations, we propose a theoretically grounded method for sampling from the posterior of infinite-dimensional Bayesian linear inverse problems based on amortized conditional SDMs. In particular, we prove that one of the most successful approaches for estimating the conditional score in finite dimensions - the conditional denoising estimator - can also be applied in infinite dimensions. A significant part of our analysis is dedicated to demonstrating that extending infinite-dimensional SDMs to the conditional setting requires careful consideration, as the conditional score typically blows up for small times, contrarily to the unconditional score. We conclude by presenting stylized and large-scale numerical examples that validate our approach, offer additional insights, and demonstrate that our method enables large-scale, discretization-invariant Bayesian inference.

Paper

References (62)

Scroll for more · 38 remaining

Similar papers

Peer review

Reviewer XSPZ7/10 · confidence 2/52023-06-26

Summary

- The study proposes a method to learn the posterior distribution in infinite-dimensional Bayesian linear inverse problems using amortized conditional Score-based Diffusion Models (SDMs). This extends conditional SMDs into the infinite-dimensional function space setting, as existing conditional SDMs have previously only dealt with finite-dimensional vector spaces (noting also that _unconditional_ SDMs have recently been extended to infinite dimensional vector spaces by Pidstrigach et. al). This leads the way for applications in, for example PDEs, where the unknown parameters to be estimated take the form of functions. - The key technique underlying their approach is to define the _conditional_ score in an infinite dimensional setting, extending the method of Pidstrigach et. al, who defined the _unconditional_ score in the infinite dimensional setting. - Using their definition of the conditional score in infinite dimensions, this allows them to avoid having to solve a potentially expensive proximal optimization step, as was done by Pidstrigach et. al. - The authors then provide a comprehensive theoretical analysis of the use of their conditional score in SDMs and show: - How this newly defined score is used as a reverse drift of the diffusion process, which leads to a generative model that samples from the correct target conditional distribution under certain conditions. - That as long as you start from the invariant distribution of the diffusion process, the reverse SDE converges to the target distribution exponentially fast - By explicitly computing the expected square norm of the conditional score, they shows that a uniform in time estimate is not always true for the conditional score. This leads them to provide a set of conditions to be satisfied to ensure a uniform in time estimate for a general class of prior measures. - That the conditional score can be estimated via a conditional denoising score matching objective in infinite dimensions. - That the conditional denoising estimator is a consistent estimator of the conditional score in infinite dimensions. - That unlike the unconditional score, for noiseless observations the conditional score blows up as T->0 - They present a small toy experiment that validates their approach by demonstrating the applicability of their method in approximating non-Gaussian multi-modal distributions.

Strengths

I will address strengths and weaknesses across the four dimensions (Originality, Quality, Clarity, Significance) below.

Weaknesses

**Originality:** - Are the tasks or methods new? - Yes, the authors make a novel contribution to the quickly-growing diffusion model literature, by showing how to extend conditional SDMs into the infinite-dimensional vector space, thus opening the door to a wider range of applications (PDEs, etc) - Is the work a novel combination of well-known techniques? Is it clear how this work differs from previous contributions? - The authors are very clear about: 1. How they are starting with the recently-proposed framework of Pidstrigach et al 2. The specific point where they deviate from, and then extend, Pidstrigach's work (Definition 2, eq. 8) - Further, they provide a comprehensive analysis of the use of their defn 2 in SDMs in sections 4 and 5. - However I should state that I am not adequately familiar with the mathematical techniques deployed in this paper to check their technical claims for correctness or novelty, so please differ to another reviewer with more expertise in this area. - Is related work adequately cited? - Yes. **Quality:** - Is the submission technically sound? - It appears to be, although as previously stated I am not sufficiently versed in their mathematical techniques to be 100% sure. - Are claims well supported (e.g., by theoretical analysis or experimental results)? - Theoretically yes, experimentally no. I realize this is a theory paper and don't necessarily expect full-scale experiments on PDEs, but in Section 6 I expected to see at least 2 obvious baselines which were foreshadowed in the body of the text but not experimentally validated on the toy experiments. These are: 1. How does their method compare to the crude, discretization-based approach discussed in lines 56 - 61: > A straightforward solution may be to discretize the infinite-dimensional input and output function spaces into finite-dimensional vectors, and apply SDMs to learn the posterior. Yet theoretical studies of current DMs suggest that performance guarantees do not generalize well on increasing dimensions [7, 9, 33]. This is precisely why Stuart’s guiding principle to study a Bayesian inverse problem for functions— “avoid discretization until the last possible moment” [41] — is more than ever critical to the use of SDMs. A primary motivation for their method is "Stuart's Principle", which says to avoid discretization until the last possible moment. I would have liked to see experimentally why this principle is so important on the simple toy examples they have provided. 2. How does their method compare to the approach proposed by Pidstrigach for conditional sampling, which uses a proximal optimization step? This is a second primary motivation for their approach (and appears in the abstract) - their method is more performant because it avoids solving an optimization problem at each timestep. As far as I can tell it has not been experimentally validated that their approach is faster or more performant than the baseline method of Pidstrigach. Evidence showing their approach either gets better samples, or get samples of the same quality but more efficiently, is needed, given that a primary motivation for their approach is that the baseline method of Pidstrigach may be too computationally costly because of their use of proximal optimization. - Are the methods used appropriate? - Yes - Is this a complete piece of work or work in progress? - Yes, modulo the missing baselines discussed above. I believe such baselines are needed to consider this a complete piece of work, given how "Stuart's Principle" and "avoid proximal optimization" play a key role in the storyline and motivation for their technique. - Are the authors careful and honest about evaluating both the strengths and weaknesses of their work? - Yes. In particular I found their observation, that unlike the unconditional score, for noiseless observations the conditional score blows up as t->0, particularly interesting. However in the conclusion I would have liked to read more about the authors' reflections on the strengths/weaknessness/future directions of their approach, both from a technical standpoint and from the standpoint of potential downstream applications of this work. **Clarity:** - Is the submission clearly written? - I found all writing up until section 4 relatively easy to follow. I began to get lost around sections 4 and 5 and couldn't follow the math, but I attribute this largely to not being comfortable with stochastic differential equations. I could still follow the high-level plot in these sections, but will defer to other reviews to evaluate the technical claims. - Is it well organized? - Yes - Does it adequately inform the reader? - Yes **Significance:** - Are the results important? - Yes, although I would have liked the authors to motivate the applications of their approach more thoroughly. I believe they only listed PDEs as an example for why you would want to use this approach but surely there are more applications than just PDEs, no? - Are others (researchers or practitioners) likely to use the ideas or build on them? - Yes - Does the submission address a difficult task in a better way than previous work? - Theoretically it appears so, but baselines are needed to experimentally validate these claims. - Does it advance the state of the art in a demonstrable way? - Theoretically it appears so, but baselines are needed to experimentally validate these claims. - Does it provide unique data, unique conclusions about existing data, or a unique theoretical or experimental approach? - Yes, they build on the work of Pidstrigach in a novel way.

Questions

- What are some other applications of this work besides PDEs? Can you include a few of them in the introduction so the reader doesn't think applications of your work is limited only to PDEs? - In the conclusion the authors say their method "is able to perform conditional sampling directly on infinite dimensional Hilbert spaces." Don't Gaussian processes also allow you to perform conditional sampling directly on infinite dimensional Hilbert spaces? What are the differences between your approach and Gaussian processes? When would I want to use one vs. the other? Could a GP be used as a baseline in your Figure 1? If so it would be very interesting to see how it compares. - The authors also say "We also show that the conditional score can have a singular behavior at small times when the observations are noiseless, in contrast with the unconditional score under similar hypotheses.". Is this just a limiting phenomenon of theoretical interest, or do you expect this to cause difficulties in practice? Are there any applications where we should expect observations to be noiseless or does that assumption never hold in practice?

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

2: You are willing to defend your assessment, but it is quite likely that you did not understand the central parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

- There does not appear to be a "Limitations and Broader Impacts" statement in this work.

Reviewer Q2NB8/10 · confidence 3/52023-06-29

Summary

This paper mathematically examines linear inverse problems in infinite dimensional vector spaces. Particularly, it is proved that the conditional denoising estimator is a consistent estimator of the conditional score in infinite dimension.

Strengths

The consistency of the conditional denoising estimator in infinite dimensional vector spaces is mathematically shown. For a specific case of Gaussian prior, the forward-reverse SDEs are solved exactly, which shows an exponentially fast convergence in the reverse SDE. A sufficient condition for the success of the score-based diffusion model framework is presented for the infinite dimensional version.

Weaknesses

The considered inverse problems are not of the plug-and-play type, which would limits the practical utility. Numerical example is limited to a one-dimensional toy model.

Questions

Gaussian process (GP) is a popular method for inverse problem in Hilbert space. I wonder if the current problem would have a certain connection to GP, in particular, in the case of Gaussian prior discussed in section 4.

Rating

8: Strong Accept: Technically strong paper, with novel ideas, excellent impact on at least one area, or high-to-excellent impact on multiple areas, with excellent evaluation, resources, and reproducibility, and no unaddressed ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

Descriptions on the motivation to examine the inverse problem in Hilbert space are lacking. In what practical situations do inverse problems in Hilbert space come out? For instance, GP is widely used for Bayes optimization, which is an optimization scheme for black-box functions. Additional writing about possible applications of the inverse problem in Hilbert space would make the paper more attractive.

Reviewer WxYS6/10 · confidence 3/52023-07-05

Summary

This paper proposed a method to deal with inverse problems in infinite dimensions using conditional-score-based models. Specifically, they propose to directly learn the posterior distribution in infinite-dimensional Bayesian linear inverse problems using amortized conditional SDMs. Moreover, this paper also discussed the robustness of the learned distribution against perturbations of the observations. A numerical experiment is conducted to validate the efficiency.

Strengths

1. This paper proposed an interesting method to deal with infinite-dimensional Bayesian linear inverse problems. 2. It provides a detailed analysis of the forward-reverse conditional SDE framework in the case of a Gaussian prior measure. 3. It provides a set of conditions to ensure a uniform in-time estimate for a general class of prior measures.

Weaknesses

1. Regarding the introduced definition of the conditional score and the result that the conditional score can be estimated via a conditional denoising score matching, it seems that they are straightforward extensions of the unconditional case. For score-based models, the key is to learn a general distribution using score matching, regardless of whether it is a conditional distribution or a non-conditional distribution. I mean, there is actually no fundamental difference between a conditional distribution and non-conditional distribution, i.e., score-based matching applies to them or other general distributions equally. 2. From my understanding, and also as suggested in the abstract, this paper only focuses on infinite-dimensional Bayesian linear inverse problems, rather than arbitrary inverse problems. The title of this paper is kind of misleading. On the other hand, for linear inverse problems, the score of the likelihood term is relatively easy to obtain, compared to directly training a conditional score network. Please correct me if I am wrong.

Questions

Some additional questions: 1. What is the difference between (2) and (3)? 2. The experiment part only considers a simple low-dimensional problem, is it possible to add some results of large-scale problems?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

2 fair

Limitations

See above.

Reviewer hg8t8/10 · confidence 3/52023-07-06

Summary

Score-based diffusion models are successful in solving inverse problems in a finite-dimensional setting, but infinite-dimensional diffusion models needs to be constructed with care, as the definitions for Lebesgue measures and densities become less clear. The authors extends the work of Pidstrigach et al. [33] of unconditional score matching in infinite dimensions to a conditional setting.

Strengths

I recommend a strong accept for the paper because of its theoretical soundness and its approachable presentation in its explanation. The paper presents a theoretically elegant and principled approach to solving inverse problems in infinite dimensions, as it is guided by Stuart’s principle and does not involve projection to finite vector spaces and discretization when unnecessary. The paper also analyzes a general scenario with prior distributions absolutely continuous w.r.t. the Gaussian measure, and propose analogous results to Pidstrigach et al. [33].

Weaknesses

I cannot identify a specific point of weakness that has to be addressed. One can argue against its simple experiment but I see a proof-of-concept experiment sufficient for this paper.

Questions

I have a few questions regarding the general aspects of Bayesian inference in infinite dimensions, as I am not familiar with the exact formulation. - Observational model in the paper occupies a finite-dimensional subspace. Is there a scenario where one cannot find an orthonormal basis such that the observation $y$ only spans a finite subspace? - While I understand Assumption 1 _can_ be satisfied under certain conditions given by the prior measure's Radon-Nikodym derivatives, but does this assumption suffer from finite training data? If we think about extreme settings with only a few training data, the score matching essentially tries to memorize these noiseless data points, causing an automatic violation of the assumption.

Rating

8: Strong Accept: Technically strong paper, with novel ideas, excellent impact on at least one area, or high-to-excellent impact on multiple areas, with excellent evaluation, resources, and reproducibility, and no unaddressed ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

4 excellent

Presentation

3 good

Contribution

3 good

Limitations

- Line 15 typo: it should read "extension of ... to the *conditional* setting". - Minor citation error on Line 72: The seminal score matching paper [17] has Hyvärinen as the sole author. - Line 293 typo: “… proposition _on_ such set of conditions”

Reviewer pjkQ6/10 · confidence 4/52023-07-07

Summary

The author extends score based diffusion from finite dimensional processes to separable Hilbert space processes. They demonstrate that on a non-linear toy data set that the method can work.

Strengths

The paper reads very well and is easy to follow. It is important to study what happens in general separable Hilbert spaces as many algorithms and models breaks down in the infinite dimensional setting. And even though you always do real application in the finite dimensional setting it is still important because if the method works in a separable Hilbert space setting it will not break down when you increase the precision your finite dimensional discretization.

Weaknesses

The numerical experiment is not very convincing. The results for this toy problem are not very impressive. I suspect that if one runs multiple parallel version of a Crank–Nicolson algorithm one would get better result. It also would be interesting to see how the numerical method scale in practice when you increase the resolution of the grid.

Questions

-When you in section 4. let $(Af)_i=(v_k,f)$ where $v_k$ is a eigenvector, is it not very obvious what you get? I mean the problem is a infinite series of independent processes and the data don't couple the processes. Does it not mean that you are back to the finite dimensional case (since if $v_j \notin A$ then for that $j$ the processes is equivalent to the prior? -In the application setting I don't get the prior on $X_0$ and when I look in the code it looks one has used $x_0$ as just a line? - What is the operator $C$ in this example? very very minor: Do you really have to write infinite dimensional in each sentence?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

2 fair

Limitations

.

Reviewer hg8t2023-08-10

Post-rebuttal comment

I thank the authors for addressing the points I laid out in the review, and maintain the same score assessment for this paper.

Reviewer XSPZ2023-08-11

Thank you for your response

Thank you for your response, which I found convincing. Given the new (and impressive!) experiments, I've happily increased my score from 6->7.

Reviewer Q2NB2023-08-17

Thank you for the reply. I keep my evaluation as it is.

Reviewer WxYS2023-08-19

Thanks for the rebuttal.

Thank the authors' detailed responses, and I increased my score accordingly.

Program Chairsdecision2023-09-21

Decision

Accept (spotlight)

© 2026 NYSGPT2525 LLC