Pith. sign in

REVIEW 4 major objections 5 minor 62 references

Diffusion Model Guided Sampling with Pixel-Wise Aleatoric Uncertainty Estimation

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Pixel-wise score variance doubles as a diffusion model's quality map and steering signal.

desk verdict Genuinely new training-free uncertainty estimator for diffusion sampling, but the theoretical curvature claim does not survive contact with Algorithm 1. read the letter →

arxiv 2412.00205 v1 pith:SFHQCV7U submitted 2024-11-29 cs.CV cs.LGstat.ML

Michele De Vita , Vasileios Belagiannis This is my paper · ORCID
classification cs.CVcs.LGstat.ML
keywords diffusionmodelsaleatoricuncertaintypixel-wiseestimationuncertainty-guidedsamplingscore-basedgenerativemodelingFisherinformationFIDDDIM
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 proposes a training-free way to read pixel-wise aleatoric uncertainty out of a diffusion model while it samples: at each denoising step, predict the clean image, renoise it $M$ times, and take the variance of the model's denoising scores across those perturbed inputs. It then shows that, under regularity conditions, this variance equals the expected second derivative of the noising distribution's log-density, so each uncertainty map is a curvature map of the diffusion process. The same map guides sampling: pixels above a percentile threshold get their scores pushed along the gradient of the uncertainty, and the authors report that this improves FID by about one point on ImageNet and CIFAR-10 and also filters out low-quality samples. The practical value is that diffusion models gain a cheap, quantitative quality signal and a steering mechanism without retraining or extra networks.

What carries the argument

The load-bearing object is Algorithm 1's uncertainty map: starting from the current noisy image $X_t$, compute the DDIM predicted clean image $\hat{X}_0$, draw $M$ noisy samples $\hat{X}_t^i \sim q(X_t|\hat{X}_0)$, evaluate the denoising network on each, and set $U_t$ to the diagonal variance of the stacked scores. Algorithm 2 turns that map into a guidance update $\hat{\varepsilon}_t = \varepsilon_t + \lambda\,\mathbb{I}[U_t > p]\cdot \partial U_t/\partial \varepsilon_t$, where $\mathbb{I}$ is the indicator of pixels above the $p$-th percentile threshold and $\lambda$ is the update strength. The identity connecting the two is Eq. 9, the Fisher-information relation between score variance and curvature; it licenses reading $U_t$ as second-order information about the noising distribution and gives the guidance update its stated rationale.

What would settle it

On a Gaussian mixture with a known closed-form score, compute the exact marginal Fisher information $-\mathbb{E}[\partial^2\log q_t(X_t)/\partial X_t^2]$ and compare it with Algorithm 1's $U_t$ at the same time steps. If $U_t$ deviates systematically from the marginal curvature, especially when the DDIM reconstruction $\hat{X}_0$ lands far from any clean sample, the Eq. 9 interpretation is refuted.

Watch

Extended reading notes

Core claim

The central claim is that the diagonal variance of the scores $\varepsilon_\theta(\hat{X}_t^i,t)$ over $M$ Monte-Carlo renoised versions of the DDIM-predicted clean image, $U_t = \mathrm{diag}(\mathrm{Var}_i[\varepsilon_\theta(\hat{X}_t^i,t)])$, approximates the expected outer product of the score, $\mathbb{E}[\nabla_{X_t}\log q_t(X_t)\,\nabla_{X_t}\log q_t(X_t)^\top]$, and by the Fisher-information identity this equals the negative expected Hessian $-\mathbb{E}[\partial^2 \log q_t(X_t)/\partial X_t\partial X_t^\top]$. In the paper's own terms, $U_t$ is therefore a Monte-Carlo estimate of the curvature of the marginal noising distribution $q_t(X_t)=\int p_{\mathrm{data}}(X_0)q(X_t|X_0)\,dX_0$, with high values marking pixels the model finds hard to reconstruct consistently. The authors further claim that gradient ascent on $U_t$ at pixels above a percentile threshold steers the reverse process toward higher-quality images, and their experiments report consistent FID gains across resolutions and model families.

Load-bearing premise

The identity in Eq. 9 is argued for the marginal noising distribution $q_t(X_t)$, but Algorithm 1 computes its variance from samples of the conditional distribution $q(X_t|\hat{X}_0)$ built from one DDIM-predicted clean image, and the paper gives no argument that the plug-in conditional variance equals the marginal Fisher information.

Editorial extensions

If this is right

  • A diffusion model can report where it is uncertain at the final denoising steps for 20 extra score evaluations in the paper's setup ($M=5$ over four steps), compared with about 130 for the existing pixel-wise Bayesian baseline.
  • Filtering the generated pool using the uncertainty score improves FID over random selection on ImageNet and CIFAR-10, providing a practical low-quality filter.
  • Guiding the sampler on high-uncertainty pixels improves FID by about one point across ADM, U-ViT, and DDPM at resolutions from 64 to 512.
  • The method transfers across samplers such as DDIM and second-order DPM solvers, and to text-to-image models, with qualitative results showing fewer artifacts.
  • Uncertainty concentrates in foreground elements and in the late denoising steps, so the maps localize the pixels the model struggles to synthesize.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • We infer that the curvature reading depends on the plug-in conditional $q(X_t|\hat{X}_0)$ standing in for the true marginal $q_t(X_t)$; a direct comparison of $U_t$ with the exact Fisher information on a synthetic density would be the natural test.
  • The update rule differentiates $U_t$ with respect to the score $\varepsilon_t$, a heuristic move not derived from Eq. 9; ablating it against a simpler high-uncertainty score scaling would isolate what the gradient term contributes.
  • If the curvature interpretation survives that test, the same map could support other uses not explored here, such as artifact localization, adaptive compute allocation, or quality control in safety-critical generation.
  • The reported FID gains are about one point, and the appendix shows that large guidance strengths or low percentiles change scene content, so the working range of $p$ and $\lambda$ needs calibration per model.
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

4 major / 5 minor

Summary. The paper proposes a training-free method to estimate pixel-wise aleatoric uncertainty during diffusion-model sampling. At each denoising step the algorithm builds a DDIM estimate X̂0 from the current state Xt, draws M Gaussian-perturbed samples around X̂0, runs the score network on these samples, and defines the uncertainty map as the diagonal of the sample variance of the resulting scores. The authors claim that this uncertainty approximates the curvature of the noising distribution, specifically the expected second derivative of log q(Xt), and they use the uncertainty map to guide sampling by applying gradient ascent to high-uncertainty pixels. Experiments on ImageNet and CIFAR-10 with ADM, U-ViT, DDPM, and Stable Diffusion compare the method with BayesDiff and MC-Dropout for low-quality-sample filtering, reconstruction uncertainty metrics (AUSE/AURG), and FID under guidance.

Significance. The idea of a training-free, scheduler-agnostic uncertainty signal that can both filter samples and steer generation is attractive, and the paper's strengths include a public code release, a broad set of models and datasets, and a favorable NFE comparison (20 vs. roughly 130 for BayesDiff). If the curvature interpretation were valid, the method would offer a principled and cheap way to improve sampling quality. However, the central theoretical equivalence is not established: the Fisher-information identity in Eq. (9) is proven for the marginal q_t, while Algorithm 1 estimates a conditional plug-in variance. In addition, the empirical gains are modest and inconsistent: BayesDiff attains a better FID on ImageNet256 in Table 1, and the guided-sampling improvements in Table 3 are about one FID point without error bars or multiple seeds. As it stands, the contribution is an ungrounded heuristic with limited evidence of practical benefit.

major comments (4)
  1. [Sec. 3.5, Eq. (9), Appendix A, Algorithm 1] The main theoretical claim is not established. Appendix A proves the standard Fisher-information identity E_q[s s^T] = -E_q[∂² log q(s)] for draws from the marginal distribution q_t(X_t), but Algorithm 1 computes the sample variance of scores at points X̂_i^t drawn from the conditional Gaussian q(·|X̂0), where X̂0 is the DDIM point estimate from Eq. (7). No importance weighting, correction term, or additional argument shows that the plug-in conditional variance equals the marginal Fisher information. This is not a technicality: for a mixture such as q_t = 0.5 N(-1,1) + 0.5 N(1,1), the variance of the score under one conditional component can differ substantially from the marginal Fisher information. The centered sample variance in Eq. (8) also coincides with the uncentered second moment in Eq. (9) only if the mean score under the sampling distribution is zero, which is not guaranteed for the conditional plug-in. Consequently, the equivalence between U_t and the curvature of the noising distribution is unsupported, and the rationale for the guidance update in Eq. (10) collapses.
  2. [Sec. 3.6, Eq. (10), Algorithm 2] The guidance mechanism is not justified. The update ε̂_t = ε_t + λ(I[U_t > p] · ∂U_t/∂ε_t) treats ∂U_t/∂ε_t as a gradient-ascent direction, but U_t depends on ε_t through the DDIM estimate X̂0 in Eq. (7) and then through the M perturbed samples in Algorithm 1. The paper does not specify how this gradient is computed, nor does it show that stepping ε_t along this direction increases the curvature of q_t as claimed. The sentence in Sec. 3.6 that maximizing the second derivative improves the convergence rate and cites [33] is not supported by that reference, which is a statistics textbook rather than an optimization or diffusion result. As written, the guidance step is a heuristic whose stated theoretical motivation is not valid.
  3. [Table 3 and Table 1] The empirical evidence for the central practical claims is weak. In Table 3, the FID improvements from uncertainty guidance are about 0.9–1.1 points on 10,000 generated samples, with no standard errors, confidence intervals, or multiple seeds; FID differences of this size are within the sampling noise of the metric. In Table 1, the proposed filtering method is not uniformly better than the baseline: BayesDiff achieves a clearly lower FID on ImageNet256 (6.81 vs. 7.80), and several of the reported improvements are very small (e.g., 3.289 vs. 3.254 on ImageNet64). The abstract's claim of 'state-of-the-art performance' is therefore not supported by the reported numbers, and the guided-sampling result lacks a comparison to a similarly priced baseline with the same compute budget.
  4. [Sec. 4.1 and Fig. 10] The guidance results depend on several hand-chosen hyperparameters—M = 5, p = 95, λ = 1.0, and the uncertainty timestep window (steps 45–48)—but no sensitivity analysis is provided for the ImageNet or CIFAR-10 benchmarks. The only ablation (Fig. 10) is qualitative and on Stable Diffusion, where changes in p and λ can substantially alter image content. Since the method is proposed as a general guidance technique, the lack of ablations on the benchmark datasets makes it unclear whether the small FID gains are robust or are the result of tuned settings.
minor comments (5)
  1. [Sec. 3.2, Eq. (2)] The reparametrization is written as X_t = √ᾱ_t X0 + (1-ᾱ_t) ε; it should be X_t = √ᾱ_t X0 + √(1-ᾱ_t) ε. Equation (1) also appears to define the conditional distribution of X_t given X_{t-1}, not X_{t-1} given X_t, as the notation suggests.
  2. [Sec. 4.2] The text says the uncertainty estimation results are presented in Table 4, but the results described (filtering FID and comparisons with BayesDiff/MC-Dropout) appear in Table 1; Table 4 in the appendix reports Precision/Recall. Please correct the cross-reference.
  3. [Algorithm 2 and Sec. 4.1] The use of the percentile threshold is ambiguous. Algorithm 2 uses a per-image threshold 'percentile(U_t, p)', while Sec. 4.1 describes a global 95-th percentile 'of the uncertainty computed over the 10,000 samples'. Please clarify which implementation was used and whether p is the same in both places.
  4. [Appendix A] The proof is stated for a scalar x and assumes conditions (15)–(16), but the paper uses vector-valued image tensors and a diagonal of the full covariance matrix. The vector case and the boundary decay of ∂q/∂x at infinity are not discussed, so the proof as written does not directly cover the setting of the main text.
  5. [Throughout] There are numerous presentation issues: 'threhsolds' in Algorithm 2, 'V AEs' in the Related Work, inconsistent capitalization such as 'Imagenet64' vs. 'ImageNet64' and 'U-VIT' vs. 'U-ViT', and Table 1's caption says 'with and without uncertainty guidance' when the experiment is actually low-quality filtering.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation found: Eq. 9 is an external Fisher-information identity, and the uncertainty-guided sampling is evaluated on external benchmarks with fixed hyperparameters rather than fitted to the target FID.

full rationale

The paper's claimed derivation chain is not circular. The uncertainty map U_t is defined in Eq. 8 and Algorithm 1 as the sample variance of score-network outputs over perturbations drawn from the conditional noising distribution q(X_t|X_hat_0), where X_hat_0 is the DDIM point estimate from Eq. 7. Equation 9 then invokes the standard Fisher-information identity E[s s^T] = -E[∂^2 log q], citing external references [12,33]; the identity is not manufactured by the paper and its proof in Appendix A is a textbook calculation. The load-bearing assumption that the plug-in conditional variance equals the marginal Fisher information is not proved, and the guidance update in Eq. 10 is therefore heuristically motivated rather than forced by a self-referential construction. However, an unsupported equivalence is a correctness gap, not a circularity: no output quantity is defined in terms of the claim it is supposed to establish, and no parameter is fitted to the FID scores it later 'predicts.' The reported FID improvements in Table 3 are external benchmark comparisons with fixed M=5, p=95th percentile, and lambda=1.0, so the empirical claim is not an artifact of fitting. The paper's self-citations (e.g., Hornauer and Belagiannis works in the related-work section) are not load-bearing for the curvature derivation. Hence there is no circular step to exhibit.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central derivation rests on four hand-chosen hyperparameters (M=5, p=95%, lambda=1.0, and a timestep window) and on an unproven substitution: re-noised samples from a predicted clean image are treated as draws from the true marginal noising distribution. The regularity conditions for the Fisher identity are standard but unverified; the sensitivity-to-uncertainty and curvature-to-quality links are imported heuristics. No invented entities are introduced.

free parameters (4)
  • M (number of Monte Carlo perturbed samples) = 5 (20 in Appendix Fig. 11)
    Number of re-noised samples used to compute the variance in Algorithm 1; chosen in Sec. 4.1 and affects both the uncertainty estimate and the computational cost shown in Tables 5 and 6.
  • p (uncertainty percentile threshold) = 95th percentile
    Mask threshold in Algorithm 2; set in Sec. 4.1 Implementation Details, and Appendix Fig. 10 shows changing p alters scene content substantially.
  • lambda (guidance strength) = 1.0
    Update strength in Eq. 10; chosen in Sec. 4.1, and Appendix Fig. 10 shows increasing lambda can fundamentally change the scene structure.
  • uncertainty timestep window = steps 45-48 of 50
    Pixel-wise uncertainty is summed only over these steps; the choice is motivated by Fig. 3 but is a modeling decision that affects the quality filtering results.
assumptions (6)
  • standard math The noising distribution q_t(X_t) = ∫ p_data(X0) q(X_t|X0) dX0 is twice continuously differentiable with finite expected squared second derivative of its log-density (regularity conditions in Appendix A, Eq. 15-16).
    Needed for the Fisher information identity in Eq. 9; standard regularity, but asserted for all t and all data distributions without verification.
  • domain assumption The trained score network epsilon_theta(X_t,t) equals the true score ∇ log q_t(X_t) at the re-noised inputs X_i^t.
    Score matching trains against conditional scores of the noising process; using the network at perturbed inputs as a proxy for the marginal score is assumed throughout Sec. 3.5.
  • ad hoc to paper Monte Carlo draws X_i^t from q(X_t | X0_hat), where X0_hat is the DDIM estimate in Eq. 7, are representative samples from the marginal q_t(X_t).
    Algorithm 1 lines 2-4 define this plug-in sampling; without it Eq. 9 does not follow, and no correction is given.
  • domain assumption Sensitivity of model outputs to input perturbations is a valid proxy for aleatoric uncertainty, as established by Mi et al. [35].
    The paper imports this equivalence and never validates it for diffusion score networks.
  • ad hoc to paper Performing gradient ascent on the uncertainty map, equivalently on the second derivative of log q_t(X_t), improves the perceptual quality of generated samples.
    Sec. 3.6 cites optimization convergence intuition from [33], but no theorem connects higher Fisher information of the noising distribution to lower FID.
  • domain assumption The variance of M=5 score estimates is an unbiased and sufficient estimate of the expectation in Eq. 9.
    No bias or convergence analysis is supplied for such a small Monte Carlo sample.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Diffusion Model Guided Sampling with Pixel-Wise Aleatoric Uncertainty Estimation." pith.science (2026). https://pith.science/paper/SFHQCV7U

@misc{pith2026241200205,
  author       = {Pith},
  title        = {Pith review of: Diffusion Model Guided Sampling with Pixel-Wise Aleatoric Uncertainty Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SFHQCV7U}},
  note         = {Machine review of arXiv:2412.00205}
}
read the original abstract

Despite the remarkable progress in generative modelling, current diffusion models lack a quantitative approach to assess image quality. To address this limitation, we propose to estimate the pixel-wise aleatoric uncertainty during the sampling phase of diffusion models and utilise the uncertainty to improve the sample generation quality. The uncertainty is computed as the variance of the denoising scores with a perturbation scheme that is specifically designed for diffusion models. We then show that the aleatoric uncertainty estimates are related to the second-order derivative of the diffusion noise distribution. We evaluate our uncertainty estimation algorithm and the uncertainty-guided sampling on the ImageNet and CIFAR-10 datasets. In our comparisons with the related work, we demonstrate promising results in filtering out low quality samples. Furthermore, we show that our guided approach leads to better sample generation in terms of FID scores.

Figures

Figures reproduced from arXiv: 2412.00205 by the authors.

Figure 1
Figure 1. Visual Results I. We provide qualitative samples of uncertainty guidance applied to Stable Diffusion 3 [ [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of our uncertainty estimation algorithm for [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. We present posterior uncertainty in pixel (left) and la [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Left: generated image from DDPM trained on Ima [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Left: generated image from DDPM trained on Ima [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Additional visual results of uncertainty guidance applied to Stable Diffusion. For each pair of images, the [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Additional visual results of uncertainty guidance applied to Stable Diffusion. For each pair of images, the [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Uncertainty maps obtained from our proposed method. Coherently to our findings in the main article (Figure 3 in the main [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: Uncertainty maps obtained from MC Dropout. While our method has high uncertainty on foreground objects, we observe that [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: Hyperparameter sweep for uncertainty-guided sampling on Stable Diffusion 1.5. The top row shows the effect of varying [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]
Figure 11
Figure 11. Figure 11: Uncertainty low quality filtering as in Table 1 of the main article, but using different number of perturbated samples for [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]
Figure 12
Figure 12. Figure 12: Additional visual results of uncertainty guidance applied to Stable Diffusion. For each pair of images, the [PITH_FULL_IMAGE:figures/full_fig_p017_12.png]
Figure 13
Figure 13. Figure 13: Additional visual results of uncertainty guidance applied to Stable Diffusion. For each pair of images, the top row shows the generated image without uncertainty guidance while the bottom row shows the same image generated with uncertainty guidance. In the second colu…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 58 canonical work pages

  1. [33]

    Springer texts in statistics, Springer, New York, NY , 2

    Lehmann, E.L., Casella, G.: Theory of point estima- tion. Springer texts in statistics, Springer, New York, NY , 2. ed edn. (1998) 5, 6

    work page 1998

  2. [1]

    semanticscholar

    An, J., Cho, S.: Variational autoencoder based anomaly detection using reconstruction probability (2015), https : / / api . semanticscholar . org/CorpusID:36663713 3

    work page 2015

  3. [2]

    Stochastic Processes and their Applications 12(3), 313–326 (1982)

    Anderson, B.D.O.: Reverse-time diffusion equation models. Stochastic Processes and their Applications 12(3), 313–326 (1982). https://doi.org/https://doi.org/10.1016/0304- 4149(82)90051-5, https : / / www . sciencedirect . com / science / article / pii/0304414982900515 4

    arXiv 1982

  4. [3]

    Transactions on Machine Learning Research (2024), https:// openreview.net/forum?id=5VotySkajV 3

    Asthana, R., Conrad, J., Dawoud, Y ., Ortmanns, M., Belagiannis, V .: Multi-conditioned graph dif- fusion for neural architecture search. Transactions on Machine Learning Research (2024), https:// openreview.net/forum?id=5VotySkajV 3

    work page 2024

  5. [4]

    In: NeurIPS 2022 Workshop on Score-Based Methods (2022), https://openreview.net/forum? id=WfkBiPO5dsG 6

    Bao, F., Li, C., Cao, Y ., Zhu, J.: All are worth words: a vit backbone for score-based diffusion models. In: NeurIPS 2022 Workshop on Score-Based Methods (2022), https://openreview.net/forum? id=WfkBiPO5dsG 6

    work page 2022

  6. [5]

    In: Bach, F., Blei, D

    Blundell, C., Cornebise, J., Kavukcuoglu, K., Wier- stra, D.: Weight uncertainty in neural network. In: Bach, F., Blei, D. (eds.) Proceedings of the 32nd International Conference on Machine Learning. Pro- ceedings of Machine Learning Research, vol. 37, pp. 1613–1622. PMLR, Lille, France (07–09 Jul 2015), https : / / proceedings . mlr . press / v37 / blund...

    work page 2015

  7. [6]

    arXiv preprint arXiv:1806.05452 (2018) 1

    Chen, X., Pawlowski, N., Rajchl, M., Glocker, B., Konukoglu, E.: Deep generative models in the real- world: An open challenge from medical imaging. arXiv preprint arXiv:1806.05452 (2018) 1

    arXiv 2018

  8. [7]

    In: Proceed- ings of the IEEE/CVF conference on computer vision and pattern recognition

    Chong, M.J., Forsyth, D.: Effectively unbiased fid and inception score and where to find them. In: Proceed- ings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 6070–6079 (2020) 6

    work page 2020

Show all 62 references
  1. [8]

    In: NeurIPS (2021) 1

    Daxberger, E., Kristiadi, A., Immer, A., Eschenhagen, R., Bauer, M., Hennig, P.: Laplace redux–effortless Bayesian deep learning. In: NeurIPS (2021) 1

    2021

  2. [9]

    In: 2009 IEEE conference on computer vi- sion and pattern recognition

    Deng, J., Dong, W., Socher, R., Li, L.J., Li, K., Fei- Fei, L.: Imagenet: A large-scale hierarchical image database. In: 2009 IEEE conference on computer vi- sion and pattern recognition. pp. 248–255. Ieee (2009) 2, 6, 7

    2009

  3. [10]

    Advances in neural information processing systems 34, 8780–8794 (2021) 1, 3, 6

    Dhariwal, P., Nichol, A.: Diffusion models beat gans on image synthesis. Advances in neural information processing systems 34, 8780–8794 (2021) 1, 3, 6

    2021

  4. [11]

    In: Forty-first In- ternational Conference on Machine Learning 2

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., M¨uller, J., Saini, H., Levi, Y ., Lorenz, D., Sauer, A., Boesel, F., et al.: Scaling rectified flow transformers for high-resolution image synthesis. In: Forty-first In- ternational Conference on Machine Learning 2

  5. [12]

    University of Toronto, 2 edn

    Evans, M.J., Rosenthal, J.S.: Probability and Statis- tics: The Science of Uncertainty. University of Toronto, 2 edn. (2010) 5

    2010

  6. [13]

    Filos, A., Tigkas, P., McAllister, R., Rhinehart, N., Levine, S., Gal, Y .: Can autonomous vehicles iden- tify, recover from, and adapt to distribution shifts? In: International Conference on Machine Learning. pp. 3145–3153. PMLR (2020) 1, 6

    2020

  7. [14]

    In: international conference on machine learning

    Gal, Y ., Ghahramani, Z.: Dropout as a bayesian ap- proximation: Representing model uncertainty in deep learning. In: international conference on machine learning. pp. 1050–1059. PMLR (2016) 1, 2, 7

    2016

  8. [15]

    Communications of the ACM 63(11), 139–144 (2020) 1, 3

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., Bengio, Y .: Generative adversarial networks. Communications of the ACM 63(11), 139–144 (2020) 1, 3

    2020

  9. [16]

    https://huggingface.co/google/ddpm- cifar10-32 (2022) 6

    Google: Denoising diffusion probabilistic model (ddpm) trained on cifar-10 at 32x32 resolution. https://huggingface.co/google/ddpm- cifar10-32 (2022) 6

    2022

  10. [17]

    In: Chaudhuri, K., Sugiyama, M

    Grover, A., Ermon, S.: Uncertainty autoencoders: Learning compressed representations via variational information maximization. In: Chaudhuri, K., Sugiyama, M. (eds.) Proceedings of the Twenty- Second International Conference on Artificial Intelli- gence and Statistics. Proceed...

    2019

  11. [18]

    In: Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Con- ference, Lima, Peru, October 4–8, 2020, Proceedings, Part I 23

    Hemsley, M., Chugh, B., Ruschin, M., Lee, Y ., Tseng, C.L., Stanisz, G., Lau, A.: Deep generative model for synthetic-ct generation with uncertainty predictions. In: Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Con- ference, Lima, ...

    2020

  12. [19]

    Ad- vances in neural information processing systems 30 (2017) 6, 7

    Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., Hochreiter, S.: Gans trained by a two time-scale 9 update rule converge to a local nash equilibrium. Ad- vances in neural information processing systems 30 (2017) 6, 7

    2017

  13. [20]

    In: Guyon, I., Luxburg, U.V ., Bengio, S., Wal- lach, H., Fergus, R., Vishwanathan, S., Garnett, R

    Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., Hochreiter, S.: Gans trained by a two time-scale update rule converge to a local nash equilibrium. In: Guyon, I., Luxburg, U.V ., Bengio, S., Wal- lach, H., Fergus, R., Vishwanathan, S., Garnett, R. (eds.) Advances in Neu...

    2017

  14. [21]

    Advances in neural information pro- cessing systems 33, 6840–6851 (2020) 1, 3, 6

    Ho, J., Jain, A., Abbeel, P.: Denoising diffusion prob- abilistic models. Advances in neural information pro- cessing systems 33, 6840–6851 (2020) 1, 3, 6

    2020

  15. [22]

    In: Euro- pean Conference on Computer Vision

    Hornauer, J., Belagiannis, V .: Gradient-based un- certainty for monocular depth estimation. In: Euro- pean Conference on Computer Vision. pp. 613–630. Springer (2022) 2

    2022

  16. [23]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision

    Hornauer, J., Belagiannis, V .: Heatmap-based out- of-distribution detection. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. pp. 2603–2612 (2023) 3

    2023

  17. [24]

    In: Proceedings of the IEEE/CVF International Con- ference on Computer Vision

    Hornauer, J., Holzbock, A., Belagiannis, V .: Out-of- distribution detection for monocular depth estimation. In: Proceedings of the IEEE/CVF International Con- ference on Computer Vision. pp. 1911–1921 (2023) 2, 3

    2023

  18. [25]

    arXiv preprint arXiv:1704.00109 (2017) 3

    Huang, G., Li, Y ., Pleiss, G., Liu, Z., Hopcroft, J.E., Weinberger, K.Q.: Snapshot ensembles: Train 1, get m for free. arXiv preprint arXiv:1704.00109 (2017) 3

    2017 arXiv

  19. [26]

    In: Pro- ceedings of the European Conference on Computer Vision (ECCV)

    Ilg, E., Cicek, O., Galesso, S., Klein, A., Makansi, O., Hutter, F., Brox, T.: Uncertainty estimates and multi-hypotheses networks for optical flow. In: Pro- ceedings of the European Conference on Computer Vision (ECCV). pp. 652–667 (2018) 7

    2018

  20. [27]

    In: International conference on machine learning

    Ioffe, S., Szegedy, C.: Batch normalization: Acceler- ating deep network training by reducing internal co- variate shift. In: International conference on machine learning. pp. 448–456. pmlr (2015) 2

    2015

  21. [28]

    Advances in neural informa- tion processing systems 34, 21696–21707 (2021) 3

    Kingma, D., Salimans, T., Poole, B., Ho, J.: Varia- tional diffusion models. Advances in neural informa- tion processing systems 34, 21696–21707 (2021) 3

    2021

  22. [29]

    Kingma, D.P., Welling, M.: Auto-encoding variational bayes (2013) 1, 3, 4

    2013

  23. [30]

    arXiv preprint arXiv:2310.11142 (2023) 1, 3, 6, 7

    Kou, S., Gan, L., Wang, D., Li, C., Deng, Z.: Bayesd- iff: Estimating pixel-wise uncertainty in diffusion via bayesian inference. arXiv preprint arXiv:2310.11142 (2023) 1, 3, 6, 7

    2023 arXiv

  24. [31]

    Krizhevsky, A.: Learning multiple layers of features from tiny images. Tech. rep. (2009) 2, 7

    2009

  25. [32]

    Advances in neural information processing systems 30 (2017) 1, 2

    Lakshminarayanan, B., Pritzel, A., Blundell, C.: Sim- ple and scalable predictive uncertainty estimation us- ing deep ensembles. Advances in neural information processing systems 30 (2017) 1, 2

    2017

  26. [34]

    Lu, C., Zhou, Y ., Bao, F., Chen, J., Li, C., Zhu, J.: Dpm-solver++: Fast solver for guided sampling of dif- fusion probabilistic models (2022) 7

    2022

  27. [35]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Mi, L., Wang, H., Tian, Y ., He, H., Shavit, N.N.: Training-free uncertainty estimation for dense regres- sion: Sensitivity as a surrogate. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 36, pp. 10042–10050 (2022) 1, 3, 4

    2022

  28. [36]

    In: International Con- ference on Learning Representations (2020) 2

    Morales-Alvarez, P., Hern ´andez-Lobato, D., Molina, R., Hern ´andez-Lobato, J.M.: Activation-level uncer- tainty in deep neural networks. In: International Con- ference on Learning Representations (2020) 2

    2020

  29. [37]

    In: Pro- ceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition

    Neumeier, M., Dorn, S., Botsch, M., Utschick, W.: Reliable trajectory prediction and uncertainty quan- tification with conditioned diffusion models. In: Pro- ceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition. pp. 3461–3470 (2024) 1

    2024

  30. [38]

    In: The Twelfth International Confer- ence on Learning Representations (2024),https:// openreview.net/forum?id=8nxy1bQWTG 3

    Nielsen, B.M.G., Christensen, A., Dittadi, A., Winther, O.: Diffenc: Variational diffusion with a learned encoder. In: The Twelfth International Confer- ence on Learning Representations (2024),https:// openreview.net/forum?id=8nxy1bQWTG 3

    2024

  31. [39]

    Advances in Neural Information Processing Systems 34, 802–814 (2021) 1, 3

    Notin, P., Hern ´andez-Lobato, J.M., Gal, Y .: Improv- ing black-box optimization in vae latent space using decoder uncertainty. Advances in Neural Information Processing Systems 34, 802–814 (2021) 1, 3

    2021

  32. [40]

    In: Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition

    Poggi, M., Aleotti, F., Tosi, F., Mattoccia, S.: On the uncertainty of self-supervised monocular depth esti- mation. In: Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition. pp. 3227–3237 (2020) 7 10

    2020

  33. [41]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10684–10695 (2022) 2, 6, 7

    2022

  34. [42]

    Advances in neural information processing systems 30 (2017) 1, 3

    Saatci, Y ., Wilson, A.G.: Bayesian gan. Advances in neural information processing systems 30 (2017) 1, 3

    2017

  35. [43]

    In: Pro- ceedings of the IEEE/CVF Winter Conference on Ap- plications of Computer Vision

    Sagar, A.: Uncertainty quantification using variational inference for biomedical image segmentation. In: Pro- ceedings of the IEEE/CVF Winter Conference on Ap- plications of Computer Vision. pp. 44–51 (2022) 3

    2022

  36. [44]

    In: International Conference on Learning Representations (2020) 1, 4, 5, 7

    Song, J., Meng, C., Ermon, S.: Denoising diffu- sion implicit models. In: International Conference on Learning Representations (2020) 1, 4, 5, 7

    2020

  37. [45]

    Advances in neu- ral information processing systems 32 (2019) 4, 5

    Song, Y ., Ermon, S.: Generative modeling by estimat- ing gradients of the data distribution. Advances in neu- ral information processing systems 32 (2019) 4, 5

    2019

  38. [46]

    In: Uncertainty in Artificial Intelligence

    Song, Y ., Garg, S., Shi, J., Ermon, S.: Sliced score matching: A scalable approach to density and score estimation. In: Uncertainty in Artificial Intelligence. pp. 574–584. PMLR (2020) 5

    2020

  39. [47]

    In: In- ternational Conference on Learning Representations (2021), https://openreview.net/forum? id=PxTIG12RRHS 4, 5, 6

    Song, Y ., Sohl-Dickstein, J., Kingma, D.P., Kumar, A., Ermon, S., Poole, B.: Score-based generative model- ing through stochastic differential equations. In: In- ternational Conference on Learning Representations (2021), https://openreview.net/forum? id=PxTIG12RRHS 4, 5, 6

    2021

  40. [48]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Szegedy, C., Vanhoucke, V ., Ioffe, S., Shlens, J., Wo- jna, Z.: Rethinking the inception architecture for com- puter vision. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 2818– 2826 (2016) 7

    2016

  41. [49]

    In: Dy, J., Krause, A

    Teye, M., Azizpour, H., Smith, K.: Bayesian uncer- tainty estimation for batch normalized deep networks. In: Dy, J., Krause, A. (eds.) Proceedings of the 35th International Conference on Machine Learning. Pro- ceedings of Machine Learning Research, vol. 80, pp. 4907–4916. PMLR...

    2018

  42. [50]

    In: 2023 IEEE/RSJ International Conference on In- telligent Robots and Systems (IROS)

    Wiederer, J., Schmidt, J., Kressel, U., Dietmayer, K., Belagiannis, V .: Joint out-of-distribution detection and uncertainty estimation for trajectory prediction. In: 2023 IEEE/RSJ International Conference on In- telligent Robots and Systems (IROS). pp. 5487–5494. IEEE (2023) 2

    2023

  43. [51]

    Ad- vances in neural information processing systems 33, 4697–4708 (2020) 2

    Wilson, A.G., Izmailov, P.: Bayesian deep learning and a probabilistic perspective of generalization. Ad- vances in neural information processing systems 33, 4697–4708 (2020) 2

    2020

  44. [52]

    In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops

    Wyatt, J., Leach, A., Schmon, S.M., Willcocks, C.G.: Anoddpm: Anomaly detection with denoising diffu- sion probabilistic models using simplex noise. In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops. pp. 650–656 (June 2022) 7

    2022

  45. [53]

    ∂ ∂Xt log q(Xt) ∂ ∂Xt log q(Xt) ⊤# (14) where

    Zhai, G., Min, X.: Perceptual image quality assess- ment: a survey. Science China Information Sciences 63, 1–52 (2020) 6 11 A. Proof of the main statement In this section we provide the proof that the expected value of the variance of the scores is equivalent to the ex- pected...

    2020

  46. [54]

    First, we expand the RHS: −E ∂2 ∂x2 log q(x) = − Z q(x) ∂2 ∂x2 log q(x)dx (18)

  47. [55]

    Using the chain rule: ∂2 ∂x2 log q(x) = ∂ ∂x 1 q(x) ∂q(x) ∂x (19) Then by applying the product rule for differentiation, which states that (u · v)′ = u · v′ + v · u′ we have that = − 1 q(x)2 ∂q(x) ∂x 2 + 1 q(x) ∂2q(x) ∂x2 (20)

  48. [56]

    Substituting this back into the integral: − Z q(x) − 1 q(x)2 ∂q(x) ∂x 2 + 1 q(x) ∂2q(x) ∂x2 ! dx = Z 1 q(x) ∂q(x) ∂x 2 dx − Z ∂2q(x) ∂x2 dx

  49. [57]

    The second term becomes zero due to the property in Eq. 15 as: Z ∂2q(x) ∂x2 dx = ∂q(x) ∂x |∞ −∞ (21) Finally, considering that q(x) is a probability distribu- tion, its derivative ∂q(x) ∂x is 0 when diverging to ±∞, hence ∂q(x) ∂x |∞ −∞ = 0 (22) Now, going back to the first te...

  50. [58]

    We can multiply and divide the integrand byq(x) with- out changing the value of the integral: Z q(x) q(x) ∂q(x) ∂x 2 1 q(x) dx (24)

  51. [59]

    This can be rewritten as: Z q(x) 1 q(x) ∂q(x) ∂x 2 dx (25)

  52. [60]

    Now, we can use the following identity: 1 q(x) ∂q(x) ∂x = ∂ log q(x) ∂x (26) 12

  53. [61]

    Substituting this identity into the previous expression, we get: Z q(x) ∂ log q(x) ∂x 2 dx (27)

  54. [62]

    This is exactly the definition of the left-hand side of the original equation: E " ∂ ∂x log q(x) 2# (28) Therefore, we have shown that the right-hand side equals the left-hand side, proving the identity. B. Additional figures Figure 4. Left: generated image from DDPM trained o...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.