Pith. sign in

REVIEW 3 major objections 4 minor 60 references

Noise Conditional Variational Score Distillation

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read NCVSD shows the unconditional score of a diffusion model encodes every denoising posterior, so one distilled generative denoiser generates in one step, refines with added compute, and acts as a plug-and-play prior for inverse problems.

desk verdict A solid diffusion distillation paper that repurposes a classical Gaussian conditioning identity to train generative denoisers, with strong experiments and idealized exactness claims that need softening. read the letter →

arxiv 2506.09416 v1 pith:GIN4XXAB submitted 2025-06-11 cs.CV

Xinyu Peng , Ziyang Zheng , Yaoming Wang , Han Li , Nuowen Kan , Wenrui Dai , Chenglin Li , Junni Zou
show 1 more author
Hongkai Xiong
This is my paper · ORCID
classification cs.CV MSC 68T0768T4562F15
keywords diffusiondistillationgenerativedenoiserdenoisingposteriorsamplingvariationalscorenoiseconditionalestimatorplug-and-playinverseproblemsfew-stepimagegenerationconsistencymodels
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

NCVSD claims that a single distilled network can replace iterative diffusion sampling without giving up any of its benefits. The enabling move is a conditional-score identity: the score of the denoising posterior $q(x_0|y_\sigma)$ can be computed exactly from a pretrained unconditional score evaluated at an effective noise level, so the otherwise intractable 'data score' inside Variational Score Distillation becomes a single teacher-network call. Training a one-step generator against this score, together with an adversarial term, yields a generative denoiser that samples the denoising posterior at any noise level. From that one model the paper obtains one-step generation from pure Gaussian noise, multi-step refinement whose intermediate distributions provably stay on the target, and an asymptotically exact plug-and-play sampler for inverse problems. If the results hold, diffusion models compress into networks that generate instantly, spend extra compute only when quality matters, and solve inverse problems with roughly twenty times fewer function evaluations than current diffusion solvers.

What carries the argument

The load-bearing object is a conditional-score identity the paper calls the noise conditional score estimator (Proposition 1): restricting to the denoising posterior and pooling the two noise observations by precision gives $\nabla_{x_t}\log q(x_t|y_\sigma) = t^{-2}(D_0(y_{\sigma_{\mathrm{eff}}}, \sigma_{\mathrm{eff}}) - x_t)$ with $y_{\sigma_{\mathrm{eff}}} = (\sigma^{-2}y_\sigma + t^{-2}x_t)/(\sigma^{-2}+t^{-2})$ and $\sigma_{\mathrm{eff}} = (\sigma^{-2}+t^{-2})^{-1/2}$. This one identity carries the argument: it converts the intractable data score of the conditional VSD gradient into a call to the pretrained teacher, leaving only the model score, which an online conditional score model $D_\phi$ learns from the generator's own output. Around this core the paper assembles the auxiliary Jensen–Shannon adversarial loss, a magnitude-preserving ControlNet-style parameterization initialized from the teacher, uncertainty-weighted loss balancing, the DDIM-type latent construction of Proposition 2 that makes multi-step sampling marginal-preserving, and PnP-GD, a Split Gibbs sampler whose ULA likelihood step uses a step size that adapts to the annealing noise level.

What would settle it

On a tractable target such as a low-dimensional Gaussian mixture, compute both sides of the Proposition 1 identity exactly or by dense Monte Carlo; any mismatch falsifies the conditional-score estimator. Then distill a generative denoiser from an exact teacher score on the same target and compare its conditional samples at several noise levels with the analytic $q(x_0|y_\sigma)$; if the sample histograms deviate measurably at converged loss, the premise that the NCVSD objective drives the generator to the denoising posterior is refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that a pretrained unconditional diffusion model already contains the score functions of every denoising posterior. Proposition 1 states that when $x_0$ is observed through two independent Gaussian channels with noise levels $\sigma$ and $t$, the posterior $q(x_0|x_t, y_\sigma)$ equals the denoising posterior at the precision-weighted average $y_{\sigma_{\mathrm{eff}}} = (\sigma^{-2}y_\sigma + t^{-2}x_t)/(\sigma^{-2}+t^{-2})$ with effective noise level $\sigma_{\mathrm{eff}} = (\sigma^{-2}+t^{-2})^{-1/2}$; consequently $\nabla_{x_t}\log q(x_t|y_\sigma) = t^{-2}(\mathbb{E}[x_0|y_{\sigma_{\mathrm{eff}}}] - x_t)$, computable through Tweedie's formula with a single forward pass of the unconditional denoiser. NCVSD inserts this estimator into the Variational Score Distillation gradient, conditioning the generator on noisy data $y_\sigma$ and estimating the model score with an online conditional score model, then adds a Jensen–Shannon adversarial loss so the student can exceed the teacher. The resulting generative denoiser $G_\theta(y_\sigma, \sigma, z)$ approximates samples of $q(x_0|y_\sigma)$ across noise levels and delivers three capabilities: one-step generation at high noise, marginal-preserving multi-step sampling through a DDIM-style latent construction (Proposition 2), and asymptotically exact plug-and-play posterior sampling via a Split Gibbs sampler whose prior step is a single denoiser evaluation. Empirically, 4-step NCVSD-L reaches FID 1.76 on ImageNet-512, below 2-step sCM-XXL (1.88) and the teacher EDM2-XXL (1.81), and PnP-GD achieves best or second-best LPIPS on linear and nonlinear inverse problems at 50 function evaluations where diffusion solvers typically need about 1,000.

Load-bearing premise

The claim rests on the distilled generator actually sampling the true denoising posterior at every noise level: the combined reverse-KL and adversarial training has no convergence guarantee, and the auxiliary score model is updated only one step per generator step with no guarantee that it keeps tracking the generator's evolving output distribution.

Editorial extensions

If this is right

  • A single distilled generative denoiser spans the speed–quality trade-off without retraining: FID on ImageNet-512 improves monotonically from 2.56 at 1 step to 2.03 at 2 steps to 1.76 at 4 steps for NCVSD-L.
  • Multi-step sampling is marginal-preserving rather than heuristic: Proposition 2 guarantees $p(x_i) = q(x_i|y_\sigma)$ for every intermediate latent, so added steps refine the sample without drifting from the target denoising posterior.
  • With the adversarial term the student overtakes the teacher: 4-step NCVSD-L on ImageNet-512 (FID 1.76) beats 2-step sCM-XXL (1.88) and the teacher EDM2-XXL (1.81) at a fraction of the function evaluations.
  • PnP-GD solves linear and nonlinear inverse problems, including phase retrieval, in about 50 function evaluations with best or second-best LPIPS, roughly 20 times fewer evaluations than diffusion solvers such as DAPS and DPS that use about 1,000.
  • The noise conditional score estimator is transportable: as the paper states, any score distillation method can adopt it to distill generative denoisers from pretrained score models.

Reading between the lines

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

  • The effective-noise substitution is a general precision-pooling fact: two independent noisy observations of $x_0$ are posterior-equivalent to their precision-weighted average at the combined noise level, so the same identity suggests a recipe beyond distillation for turning any unconditional denoiser into one that conditions on extra noisy evidence.
  • The same weights act as an unconditional sampler, a conditional sampler, and a Bayesian prior; this points toward a single model handling both text-to-image generation and inverse problems, a unification the paper demonstrates piecewise but does not push.
  • Because the adversarial loss is what allows the student to exceed the teacher, the method's ceiling may be set less by the teacher's score accuracy than by GAN-style training stability and discriminator quality, so the acknowledged tuning burden becomes the likely scaling bottleneck.
  • The $\zeta$ parameter of the multi-step construction interpolates between stochastic and deterministic refinement, and the experiments fix $\zeta = 1.0$ throughout; sweeping it at inference could expose a quality-versus-diversity trade-off the paper does not explore.
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

3 major / 4 minor

Summary. The paper proposes Noise Conditional Variational Score Distillation (NCVSD), a method to distill a pretrained diffusion model into a conditional generator that approximates samples from the denoising posterior q(x0|yσ) across a range of noise levels. The key theoretical contribution is Proposition 1, a closed-form expression for the score of the denoising posterior q(xt|yσ) in terms of the unconditional score evaluated at an effective noise level. This is integrated into a VSD-style objective, with an additional adversarial loss and a carefully engineered parameterization. The trained generator supports one-step sampling, multi-step refinement (Proposition 2), and plug-and-play posterior sampling via a Split Gibbs Sampler (PnP-GD) claimed to be asymptotically exact. Experiments on ImageNet 64/512 and FFHQ-256 inverse problems show competitive FID and LPIPS with substantially fewer function evaluations than diffusion-based baselines.

Significance. If the method works as claimed, it offers a practical bridge between one-step distillation and iterative refinement, and it provides a fast, flexible posterior sampler for inverse problems. The derivation of the conditional score estimator is self-contained and correct, and the multi-step marginal construction is a clean extension of DDIM. The code is provided, and the experimental results are strong on both class-conditional generation and inverse problems. However, the paper's principal theoretical guarantees are conditional on untested assumptions about training convergence and sampler exactness, so the significance of the empirical findings is currently supported more by the experiments than by the theory.

major comments (3)
  1. [Section 3.1, Eq. (9), Algorithm 3] The NCVSD gradient is unbiased only if D_phi(x_t,t,y_sigma,sigma) accurately estimates the score of the current generator distribution p_theta(x_t|y_sigma). In Algorithm 3, D_phi is updated once per generator step and is never guaranteed to track the non-stationary distribution of G_theta. If D_phi lags, Eq. (9) is biased and G_theta need not converge to q(x0|y_sigma). Since Proposition 2 and the Section 4 exactness claim both assume mu_theta = q(x0|y_sigma), the multi-step and PnP-GD guarantees inherit this unverified assumption. The paper should either provide a two-timescale convergence analysis, or empirical tracking diagnostics (e.g., monitor the training loss of Eq. (13) on held-out generator samples, or compare D_phi with a reference score), or explicitly state that the trained generator is only a biased approximation.
  2. [Section 4, Algorithm 1] The statement that PnP-GD achieves "asymptotically exact" sampling is only true for the idealized SGS in which the prior step samples exactly from q(x0|y_sigma) and the likelihood step runs full ULA. The implemented algorithm uses a finite, pre-specified number of ULA steps with an adaptive step size gamma_sigma = C1*(beta^{-1}*C2 + sigma^{-2})^{-1}, and applies an EMA operation to the x0 samples (Algorithm 1, lines 6-8). Both the ULA discretization and the EMA introduce a bias that is not bounded by the TV-convergence argument of Vono et al. The paper should either derive a bias bound under these modifications or revise the abstract and Section 4 to state that PnP-GD is an approximate sampler.
  3. [Section 3.3, Proposition 2] The marginal-preservation guarantee requires that each intermediate draw x0 ~ mu_theta(x0|y_sigma_eff) is an exact sample from q(x0|y_sigma_eff). The training distribution of sigma in Eq. (42) does not necessarily cover the effective noise levels sigma_eff encountered during the multi-step schedule, especially at early steps where sigma_eff can be very small. The paper does not evaluate the generator's accuracy as a function of sigma_eff. Please provide a calibration analysis (e.g., FID or conditional denoising error at different sigma_eff) or restrict the multi-step claim to the range where the generator is validated.
minor comments (4)
  1. [Abstract] The abstract claims the method "outperforms teacher diffusion models" when scaling test-time compute, but Table 1 on ImageNet-64 shows 4-step NCVSD-L FID 1.53 versus EDM2-L FID 1.33 and EDM2-XL FID 1.33, so the claim is too strong as stated.
  2. [Section 5.1, Table 1] The training iteration count "32k×3" is ambiguous; state whether this means 32k generator updates plus 32k score-model and 32k discriminator updates, and explain how this is compared fairly against the single-number iteration counts of sCD and ECM.
  3. [Section 3.1] There are several typos: "direclty" should be "directly", "unconditonal" should be "unconditional", and "repectively" in Section 5.1 should be "respectively".
  4. [Section 5.2, Table 2] The claim "record-breaking LPIPS" is supported only on the FFHQ-256 dataset with a relatively small XS teacher model; the paper should state this scope explicitly in the abstract or conclusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the NCVSD derivation is self-contained, and the teacher diffusion model is an input rather than a conclusion.

full rationale

The paper's central derivation chain is self-contained. Equation (9) is obtained in Appendix A.1 from the reverse-KL objective (8) using the standard score-function identity, with the normalization term shown to vanish explicitly. Proposition 1 is proven in Appendix A.2 by completing the square in the Gaussian joint density, so the conditional score estimator (12) follows from the pretrained unconditional score model without assuming the target posterior. Proposition 2 is a DDIM-style construction whose marginal preservation is proven by induction from Gaussian identities; it does not reintroduce the training objective as an input. The generator is trained against a reverse-KL plus adversarial objective that uses real data, and the teacher D0 is an input to the distillation, not a fitted parameter derived from the reported FID/LPIPS metrics. The multi-step and PnP-GD exactness statements are explicitly conditional on the ideal scenario where mu_theta equals q(x0|y_sigma), an assumption stated in Section 4, so the downstream sampling claims do not masquerade as free predictions. Self-citations (Peng et al. 2024 as a DWT-Var baseline and Tian et al. 2024 as related work) are not load-bearing: the comparisons are against external baselines and the cited method results are empirical benchmarks, not justification of the central derivation.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The method's central identity is standard Gaussian algebra, but the training relies on strong assumptions about teacher score accuracy, generator expressiveness, and convergence of the online score model. For inverse problems, the reported numbers depend on several per-task tuned constants, including beta, ULA coefficients, and EMA decay, which are free parameters in the evaluation.

free parameters (6)
  • Noise level distribution hyperparameters (P_mean, P_std) for t = e.g., -0.8 and 1.6 for ImageNet-64; -0.4 and 1.0 for ImageNet-512
    Chosen per dataset; they control the range of noise levels over which the generative denoiser is trained.
  • EDM noise schedule parameters for sigma (sigma_max, sigma_min, rho, N) = 80.0, 0.002, 7.0, 1000 for training; 80.0, 0.002, 2.0, 50 for PnP-GD
    Define the distribution and annealing of noise levels for y_sigma; standard from EDM but chosen per task.
  • Stochasticity strength gamma = 0.414
    Controls how much extra noise is injected into y_sigma to avoid mode collapse; hand-chosen.
  • Energy strength beta in PnP-GD = 1e-4 to 4e-3 depending on task
    Treated as a tunable hyperparameter despite the theoretical definition 2*sigma_y^2; directly controls likelihood influence.
  • ULA step size coefficients C1 and C2 = C1=0.1, C2 tuned (e.g., 0.1 for super resolution and phase retrieval)
    Empirically tuned for each inverse problem; the theoretical step size is only partially derived.
  • EMA decay mu and threshold sigma_ema = mu=0.2 to 0.6, sigma_ema=0.2 for phase retrieval, infinity otherwise
    Selected by LPIPS and PSNR trade-off; EMA perturbs the asymptotic exactness of the Split Gibbs sampler.
assumptions (5)
  • domain assumption Pretrained teacher score network D0 approximates the true score of q(xt) well for all noise levels used
    The NCVSD data score estimator in Eq. (12) is only as accurate as D0; errors in the teacher propagate into the distilled generator.
  • domain assumption The generator G_theta is expressive enough and training converges to the denoising posterior q(x0|y_sigma)
    Multi-step marginals in Proposition 2 and PnP-GD exactness in Section 4 both assume mu_theta equals q(x0|y_sigma); no convergence proof is given.
  • domain assumption The online conditional score model D_phi accurately estimates the score of the evolving generator distribution
    The VSD gradient in Eq. (9) relies on the model score estimate; D_phi is updated only one step per generator step in Algorithm 3.
  • domain assumption Split Gibbs Sampler with the adopted annealing and ULA steps converges to the target posterior in the limit
    Asymptotic exactness is inherited from Vono et al., but the paper's EMA and finite ULA steps break the stated conditions.
  • standard math Gaussian convolution identities, including Tweedie and completing the square in Proposition 1
    The core theoretical result follows from standard Gaussian identities; these steps are routine.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Noise Conditional Variational Score Distillation." pith.science (2026). https://pith.science/paper/GIN4XXAB

@misc{pith2026250609416,
  author       = {Pith},
  title        = {Pith review of: Noise Conditional Variational Score Distillation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GIN4XXAB}},
  note         = {Machine review of arXiv:2506.09416}
}
read the original abstract

We propose Noise Conditional Variational Score Distillation (NCVSD), a novel method for distilling pretrained diffusion models into generative denoisers. We achieve this by revealing that the unconditional score function implicitly characterizes the score function of denoising posterior distributions. By integrating this insight into the Variational Score Distillation (VSD) framework, we enable scalable learning of generative denoisers capable of approximating samples from the denoising posterior distribution across a wide range of noise levels. The proposed generative denoisers exhibit desirable properties that allow fast generation while preserve the benefit of iterative refinement: (1) fast one-step generation through sampling from pure Gaussian noise at high noise levels; (2) improved sample quality by scaling the test-time compute with multi-step sampling; and (3) zero-shot probabilistic inference for flexible and controllable sampling. We evaluate NCVSD through extensive experiments, including class-conditional image generation and inverse problem solving. By scaling the test-time compute, our method outperforms teacher diffusion models and is on par with consistency models of larger sizes. Additionally, with significantly fewer NFEs than diffusion-based methods, we achieve record-breaking LPIPS on inverse problems.

Figures

Figures reproduced from arXiv: 2506.09416 by the authors.

Figure 1
Figure 1. The proposed generative denoisers, distilled from pretrained diffusion models, support a variety of tasks. In (a), the generative denoiser demonstrates the ability to generate diverse samples that approximate the denoising posterior distribution at arbitrary noise levels. In (b), we present the 4-step class-conditional generation results on the ImageNet-512×512 dataset. In (c), we showcase the plug-and-play probabil… view at source ↗
Figure 2
Figure 2. Sample diversity for addressing ill-poseness. Top: box inpainting with 128×128 mask. Bottom: super resolution from 16× downsampled images. consider (1) box inpainting using a center box mask of size 128×128, (2) Gaussian deblurring with kernels sized 61×61 and a standard deviation of 3.0, and (3) motion deblurring with kernels sized 61 × 61 and a std of 0.5. For nonlinear inverse problems, we consider (1) super-reso… view at source ↗
Figure 3
Figure 3. FID v.s the number of training images. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Uncurated 4-step samples generated by NCVSD-L for class-conditional ImageNet 512×512 generation. Top: class 88 (macaw); middle: class 89 (cokatoo); bottom: class 388 (giant panda). 22 [PITH_FULL_IMAGE:figures/full_fig_p022_4.png]
Figure 5
Figure 5. Figure 5: Uncurated 4-step samples generated by NCVSD-L for class-conditional ImageNet 512×512 generation. Top: class 425 (barn); middle: class 933 (cheeseburger); bottom: class 980 (volcano). 23 [PITH_FULL_IMAGE:figures/full_fig_p023_5.png]
Figure 6
Figure 6. Figure 6: Effectiveness of EMA rate in PnP-GD [PITH_FULL_IMAGE:figures/full_fig_p024_6.png]
Figure 7
Figure 7. Figure 7: Samples of PnP-GD under different EMA decay rate µ. From left to right, µ equals to 0.0, 0.2, 0.4, 0.6, 0.8. 24 [PITH_FULL_IMAGE:figures/full_fig_p024_7.png]
Figure 8
Figure 8. Figure 8: Visual comparisons for inverse problem solving. 25 [PITH_FULL_IMAGE:figures/full_fig_p025_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 55 canonical work pages

  1. [1]

    A., Salim, A., and Zhang, S

    Balasubramanian, K., Chewi, S., Erdogdu, M. A., Salim, A., and Zhang, S. Towards a theory of non-log-concave sampling: First-order stationarity guarantees for Langevin Monte Carlo . In Proceedings of Thirty Fifth Conference on Learning Theory, pp.\ 2896--2923. PMLR, 2022

    work page 2022

  2. [2]

    Bishop, C. M. and Nasrabadi, N. M. Pattern recognition and Machine Learning. Springer, 2006

    work page 2006

  3. [3]

    and Michaeli, T

    Blau, Y. and Michaeli, T. The perception-distortion tradeoff. In Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 6228--6237. IEEE, 2018

    work page 2018

  4. [4]

    T., Klasky, M

    Chung, H., Kim, J., Mccann, M. T., Klasky, M. L., and Ye, J. C. Diffusion posterior sampling for general noisy inverse problems. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=OnD9zGAGT0k

    work page 2023

  5. [5]

    Plug-and-play split Gibbs sampler: Embedding deep generative priors in Bayesian inference

    Coeurdoux, F., Dobigeon, N., and Chainais, P. Plug-and-play split Gibbs sampler: Embedding deep generative priors in Bayesian inference. IEEE Transactions on Image Processing, 33: 0 3496--3507, 2024

    work page 2024

  6. [6]

    ImageNet : A large-scale hierarchical image database

    Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L. ImageNet : A large-scale hierarchical image database. In Proceedings of 2009 IEEE Conference on Computer Vision and Pattern Recognition, pp.\ 248--255. IEEE, 2009

    work page 2009

  7. [7]

    Analysis of Langevin Monte Carlo via convex optimization

    Durmus, A., Majewski, S., and Miasojedow, B. Analysis of Langevin Monte Carlo via convex optimization. Journal of Machine Learning Research, 20 0 (73): 0 1--46, 2019

    work page 2019

  8. [8]

    Tweedie’s formula and selection bias

    Efron, B. Tweedie’s formula and selection bias. Journal of the American Statistical Association, 106 0 (496): 0 1602--1614, 2011

    work page 2011

Show all 60 references
  1. [9]

    Geng, Z., Pokle, A., Luo, W., Lin, J., and Kolter, J. Z. Consistency models made easy. In The Thirteenth International Conference on Learning Representations, 2025. URL https://openreview.net/forum?id=xQVxo9dSID

    2025

  2. [10]

    Generative adversarial nets

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. Generative adversarial nets. In Advances in Neural Information Processing Systems 27, pp.\ 2672--2680. Curran Associates, Inc., 2014

    2014

  3. [11]

    B., Ancha, S., and Roy, D

    Grosse, R. B., Ancha, S., and Roy, D. M. Measuring the reliability of MCMC inference with bidirectional Monte Carlo . In Advances in Neural Information Processing Systems 29, pp.\ 2451--2459. Curran Associates, Inc., 2016

    2016

  4. [12]

    GANs trained by a two time-scale update rule converge to a local Nash equilibrium

    Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., and Hochreiter, S. GANs trained by a two time-scale update rule converge to a local Nash equilibrium. In Advances in Neural Information Processing Systems 30, pp.\ 6626--6637. Curran Associates, Inc., 2017

    2017

  5. [13]

    Denoising diffusion probabilistic models

    Ho, J., Jain, A., and Abbeel, P. Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems 33, pp.\ 6840--6851. Curran Associates, Inc., 2020

    2020

  6. [14]

    A style-based generator architecture for generative adversarial networks

    Karras, T., Laine, S., and Aila, T. A style-based generator architecture for generative adversarial networks. In Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp.\ 4396--4405. IEEE, 2019

    2019

  7. [15]

    Elucidating the design space of diffusion-based generative models

    Karras, T., Aittala, M., Aila, T., and Laine, S. Elucidating the design space of diffusion-based generative models. In Advances in Neural Information Processing Systems 35, pp.\ 26565--26577. Curran Associates, Inc., 2022

    2022

  8. [16]

    Analyzing and improving the training dynamics of diffusion models

    Karras, T., Aittala, M., Lehtinen, J., Hellsten, J., Aila, T., and Laine, S. Analyzing and improving the training dynamics of diffusion models. In Proceedings of 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp.\ 24174--24184. IEEE, 2024

    2024

  9. [17]

    Denoising diffusion restoration models

    Kawar, B., Elad, M., Ermon, S., and Song, J. Denoising diffusion restoration models. In Advances in Neural Information Processing Systems 35, pp.\ 23593--23606. Curran Associates, Inc., 2022

    2022

  10. [18]

    Multi-task learning using uncertainty to weigh losses for scene geometry and semantics

    Kendall, A., Gal, Y., and Cipolla, R. Multi-task learning using uncertainty to weigh losses for scene geometry and semantics. In Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 7482--7491. IEEE, 2018

    2018

  11. [19]

    Consistency trajectory models: Learning probability flow ODE trajectory of diffusion

    Kim, D., Lai, C.-H., Liao, W.-H., Murata, N., Takida, Y., Uesaka, T., He, Y., Mitsufuji, Y., and Ermon, S. Consistency trajectory models: Learning probability flow ODE trajectory of diffusion. In The Twelfth International Conference on Learning Representations, 2024. URL https...

    2024

  12. [20]

    Korbak, T., Perez, E., and Buckley, C. L. RL with KL penalties is better viewed as Bayesian inference. In Findings of the Association for Computational Linguistics: EMNLP 2022, pp.\ 1083--1091. ACL, 2022

    2022

  13. [21]

    Crafting papers on machine learning

    Langley, P. Crafting papers on machine learning. In Proceedings of the 17th International Conference on Machine Learning, pp.\ 1207--1216. Morgan Kaufmann, 2000

    2000

  14. [22]

    Lee, S., Park, D., Kong, I., and Kim, H. J. Diffusion prior-based amortized variational inference for noisy inverse problems. In Proceedings of 18th European Conference on Computer Vision, pp.\ 288--304. Springer, 2025

    2025

  15. [23]

    and Song, Y

    Lu, C. and Song, Y. Simplifying, stabilizing and scaling continuous-time consistency models. In The Thirteenth International Conference on Learning Representations, 2025. URL https://openreview.net/forum?id=LyJi5ugyJx

    2025

  16. [24]

    Contrastive energy prediction for exact energy-guided diffusion sampling in offline reinforcement learning

    Lu, C., Chen, H., Chen, J., Su, H., Li, C., and Zhu, J. Contrastive energy prediction for exact energy-guided diffusion sampling in offline reinforcement learning. In Proceedings of the 40th International Conference on Machine Learning, pp.\ 22825--22855. PMLR, 2023

    2023

  17. [25]

    Diff-Instruct : A universal approach for transferring knowledge from pre-trained diffusion models

    Luo, W., Hu, T., Zhang, S., Sun, J., Li, Z., and Zhang, Z. Diff-Instruct : A universal approach for transferring knowledge from pre-trained diffusion models. In Advances in Neural Information Processing Systems 36, pp.\ 76525--76546. Curran Associates, Inc., 2023

    2023

  18. [26]

    Mammadov, A., Chung, H., and Ye, J. C. Amortized posterior sampling with diffusion prior distillation. arXiv preprint arXiv:2407.17907, 2024. URL https://arxiv.org/abs/2407.17907

    2024 arXiv

  19. [27]

    Nguyen, T. H. and Tran, A. SwiftBrush : One-step text-to-image diffusion model with variational score distillation. In Proceedings of 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp.\ 7807--7816. IEEE, 2024

    2024

  20. [28]

    OpenAI o1 system card

    OpenAI. OpenAI o1 system card. arXiv preprint arXiv:2412.16720, 2024. URL https://arxiv.org/abs/2412.16720

    2024 arXiv

  21. [29]

    Improving diffusion models for inverse problems using optimal posterior covariance

    Peng, X., Zheng, Z., Dai, W., Xiao, N., Li, C., Zou, J., and Xiong, H. Improving diffusion models for inverse problems using optimal posterior covariance. In Proceedings of the 41st International Conference on Machine Learning, pp.\ 40347--40370. PMLR, 2024

    2024

  22. [30]

    Relative entropy policy search

    Peters, J., Mulling, K., and Altun, Y. Relative entropy policy search. In Proceedings of the 24th AAAI Conference on Artificial Intelligence, pp.\ 1607--1612. AAAI Press, 2010

    2010

  23. [31]

    High-resolution image synthesis with latent diffusion models

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., and Ommer, B. High-resolution image synthesis with latent diffusion models. In Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp.\ 10674--10685. IEEE, 2022

    2022

  24. [32]

    J., and Norouzi, M

    Saharia, C., Ho, J., Chan, W., Salimans, T., Fleet, D. J., and Norouzi, M. Image super-resolution via iterative refinement. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45 0 (4): 0 4713--4726, 2022

    2022

  25. [33]

    Multistep distillation of diffusion models via moment matching

    Salimans, T., Mensink, T., Heek, J., and Hoogeboom, E. Multistep distillation of diffusion models via moment matching. In Advances in Neural Information Processing Systems 37, pp.\ 36046--36070. Curran Associates, Inc., 2024

    2024

  26. [34]

    Adversarial diffusion distillation

    Sauer, A., Lorenz, D., Blattmann, A., and Rombach, R. Adversarial diffusion distillation. In Proceedings of 18th European Conference on Computer Vision, pp.\ 87--103. Springer, 2024

    2024

  27. [35]

    Denoising diffusion implicit models

    Song, J., Meng, C., and Ermon, S. Denoising diffusion implicit models. In The Ninth International Conference on Learning Representations, 2021 a . URL https://openreview.net/forum?id=St1giarCHLP

    2021

  28. [36]

    Pseudoinverse-guided diffusion models for inverse problems

    Song, J., Vahdat, A., Mardani, M., and Kautz, J. Pseudoinverse-guided diffusion models for inverse problems. In The Eleventh International Conference on Learning Representations, 2023 a . URL https://openreview.net/forum?id=9_gsMA8MRKQ

    2023

  29. [37]

    Loss-guided diffusion models for plug-and-play controllable generation

    Song, J., Zhang, Q., Yin, H., Mardani, M., Liu, M.-Y., Kautz, J., Chen, Y., and Vahdat, A. Loss-guided diffusion models for plug-and-play controllable generation. In Proceedings of the 40th International Conference on Machine Learning, pp.\ 32483--32498. PMLR, 2023 b

    2023

  30. [38]

    and Ermon, S

    Song, Y. and Ermon, S. Generative modeling by estimating gradients of the data distribution. In Advances in Neural Information Processing Systems 32, pp.\ 11918--11930. Curran Associates, Inc., 2019

    2019

  31. [39]

    P., Kumar, A., Ermon, S., and Poole, B

    Song, Y., Sohl-Dickstein, J., Kingma, D. P., Kumar, A., Ermon, S., and Poole, B. Score-based generative modeling through stochastic differential equations. In The Ninth International Conference on Learning Representations, 2021 b . URL https://openreview.net/forum?id=PxTIG12RRHS

    2021

  32. [40]

    Consistency models

    Song, Y., Dhariwal, P., Chen, M., and Sutskever, I. Consistency models. In Proceedings of the 40th International Conference on Machine Learning, pp.\ 32211--32252. PMLR, 2023 c

    2023

  33. [41]

    Dccm: Dual data consistency guided consistency model for inverse problems

    Tian, J., Zheng, Z., Peng, X., Li, Y., Dai, W., and Xiong, H. Dccm: Dual data consistency guided consistency model for inverse problems. In 2024 IEEE International Conference on Image Processing (ICIP), pp.\ 1507--1513. IEEE, 2024

    2024

  34. [42]

    Reward-guided controlled generation for inference-time alignment in diffusion models: Tutorial and review

    Uehara, M., Zhao, Y., Wang, C., Li, X., Regev, A., Levine, S., and Biancalani, T. Reward-guided controlled generation for inference-time alignment in diffusion models: Tutorial and review. arXiv preprint arXiv:2501.09685, 2025. URL https://arxiv.org/abs/2501.09685v1

    2025 arXiv

  35. [43]

    Split-and-augmented Gibbs sampler -- Application to large-scale inference problems

    Vono, M., Dobigeon, N., and Chainais, P. Split-and-augmented Gibbs sampler -- Application to large-scale inference problems. IEEE Transactions on Signal Processing, 67 0 (6): 0 1648--1661, 2019

    2019

  36. [44]

    ProlificDreamer : High-fidelity and diverse text-to-3D generation with variational score distillation

    Wang, Z., Lu, C., Wang, Y., Bao, F., Li, C., Su, H., and Zhu, J. ProlificDreamer : High-fidelity and diverse text-to-3D generation with variational score distillation. In Advances in Neural Information Processing Systems 36, pp.\ 8406--8441. Curran Associates, Inc., 2024

    2024

  37. [45]

    and Teh, Y

    Welling, M. and Teh, Y. W. B ayesian learning via stochastic gradient L angevin dynamics. In Proceedings of the 28th International Conference on International Conference on Machine Learning, pp.\ 681--688. Omnipress, 2011

    2011

  38. [46]

    Principled probabilistic imaging using diffusion models as plug-and-play priors

    Wu, Z., Sun, Y., Chen, Y., Zhang, B., Yue, Y., and Bouman, K. Principled probabilistic imaging using diffusion models as plug-and-play priors. In Advances in Neural Information Processing Systems 37, pp.\ 118389--118427. Curran Associates, Inc., 2024

    2024

  39. [47]

    Tackling the generative learning trilemma with denoising diffusion GAN s

    Xiao, Z., Kreis, K., and Vahdat, A. Tackling the generative learning trilemma with denoising diffusion GAN s. In The Tenth International Conference on Learning Representations, 2022. URL https://openreview.net/forum?id=JprM0p-q0Co

    2022

  40. [48]

    Consistency models improve diffusion inverse solvers

    Xu, T., Zhu, Z., He, D., Wang, Y., Sun, M., Li, N., Qin, H., Wang, Y., Liu, J., and Zhang, Y.-Q. Consistency models improve diffusion inverse solvers. arXiv preprint arXiv:2403.12063, 2024. URL https://arxiv.org/abs/2403.12063v1

    2024 arXiv

  41. [49]

    and Chi, Y

    Xu, X. and Chi, Y. Provably robust score-based diffusion posterior sampling for plug-and-play image reconstruction. In Advances in Neural Information Processing Systems 37, pp.\ 36148--36184. Curran Associates, Inc., 2024

    2024

  42. [50]

    Yin, T., Gharbi, M., Park, T., Zhang, R., Shechtman, E., Durand, F., and Freeman, W. T. Improved distribution matching distillation for fast image synthesis. In Advances in Neural Information Processing Systems 37, pp.\ 47455--47487. Curran Associates, Inc., 2024 a

    2024

  43. [51]

    T., and Park, T

    Yin, T., Gharbi, M., Zhang, R., Shechtman, E., Durand, F., Freeman, W. T., and Park, T. One-step diffusion with distribution matching distillation. In Proceedings of 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp.\ 6613--6623. IEEE, 2024 b

    2024

  44. [52]

    Freedom: Training-free energy-guided conditional diffusion model

    Yu, J., Wang, Y., Zhao, C., Ghanem, B., and Zhang, J. Freedom: Training-free energy-guided conditional diffusion model. In Proceedings of 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pp.\ 23174--23184. IEEE, 2023

    2023

  45. [53]

    Improving diffusion inverse problem solving with decoupled noise annealing

    Zhang, B., Chu, W., Berner, J., Meng, C., Anandkumar, A., and Song, Y. Improving diffusion inverse problem solving with decoupled noise annealing. arXiv preprint arXiv:2407.01521, 2024. URL https://arxiv.org/abs/2407.01521

    2024 arXiv

  46. [54]

    Adding conditional control to text-to-image diffusion models

    Zhang, L., Rao, A., and Agrawala, M. Adding conditional control to text-to-image diffusion models. In Proceedings of 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pp.\ 3813--3824. IEEE, 2023

    2023

  47. [55]

    A., Shechtman, E., and Wang, O

    Zhang, R., Isola, P., Efros, A. A., Shechtman, E., and Wang, O. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of 2018 IEEE Conference on Computer Vision and Pattern Recognition, pp.\ 586--595. IEEE, 2018

    2018

  48. [56]

    Zhao, S., Brekelmans, R., Makhzani, A., and Grosse, R. B. Probabilistic inference in language models via twisted sequential Monte Carlo . In Proceedings of the 41st International Conference on Machine Learning, pp.\ 60704--60748. PMLR, 2024

    2024

  49. [57]

    Score identity distillation: Exponentially fast distillation of pretrained diffusion models for one-step generation

    Zhou, M., Zheng, H., Wang, Z., Yin, M., and Huang, H. Score identity distillation: Exponentially fast distillation of pretrained diffusion models for one-step generation. In Proceedings of the 41st International Conference on Machine Learning, pp.\ 62307--62331. PMLR, 2024

    2024

  50. [58]

    Adversarial score identity distillation: Rapidly surpassing the teacher in one step

    Zhou, M., Zheng, H., Gu, Y., Wang, Z., and Huang, H. Adversarial score identity distillation: Rapidly surpassing the teacher in one step. In The Thirteenth International Conference on Learning Representations, 2025. URL https://openreview.net/forum?id=lS2SGfWizd

    2025

  51. [59]

    Denoising diffusion models for plug-and-play image restoration

    Zhu, Y., Zhang, K., Liang, J., Cao, J., Wen, B., Timofte, R., and Van Gool, L. Denoising diffusion models for plug-and-play image restoration. In Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp.\ 1219--1229. IEEE, 2023

    2023

  52. [60]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.