pith. machine review for the scientific record. sign in

arxiv: 2605.08144 · v1 · submitted 2026-05-02 · 💻 cs.LG · cs.AI· cs.CV

Recognition: no theorem link

NoiseRater: Meta-Learned Noise Valuation for Diffusion Model Training

Fang Wu , Haokai Zhao , Da Xing , Hanqun Cao , Tinson Xu , Yanchao Li , Xiangru Tang , Zehong Wang
show 12 more authors
Aaron Tu Kuan Pang Hanchen Wang Hongbin Lin Zeqi Zhou Yinxi Li Peng Xia Li Erran Li Molei Tao Jure Leskovec Aditya Joshi Yejin Choi
Authors on Pith no claims yet

Pith reviewed 2026-05-12 01:39 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.CV
keywords diffusion modelsmeta-learningnoise valuationtraining efficiencygenerative modelingimportance weightingbilevel optimization
0
0 comments X

The pith

A meta-learned noise rater identifies more informative noise samples to improve diffusion model training efficiency and generation quality.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper challenges the uniform treatment of noise in diffusion model training by introducing NoiseRater, a meta-learning framework that evaluates the importance of individual noise realizations. A parametric rater assigns scores based on the data sample and timestep, allowing the training objective to be reweighted adaptively. The rater itself is optimized through bilevel optimization targeting better validation performance after diffusion updates. A decoupled pipeline enables practical use by shifting from soft weighting in meta-training to hard selection during standard training. Experiments confirm that focusing on high-value noise enhances both how quickly models learn and the quality of the images they generate on datasets like FFHQ and ImageNet.

Core claim

The central discovery is that noise samples in diffusion training are not equally informative, and a meta-learned rater can value them at the instance level to prioritize those that contribute more to model improvement.

What carries the argument

The parametric noise rater that conditions on data and timestep to produce importance scores, trained via bilevel optimization for downstream validation performance.

If this is right

  • Training converges faster by focusing computational effort on informative noise samples.
  • Generated image quality improves as measured on standard benchmarks.
  • Noise valuation serves as an additional lever alongside other diffusion training techniques.
  • The two-stage pipeline allows seamless integration into existing diffusion training workflows.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar valuation approaches could be explored for other noise-based generative methods like score matching.
  • Instance-level noise selection might enable more data-efficient training regimes.
  • Dynamic rater updates during training could further adapt to the model's evolving needs.

Load-bearing premise

That the noise importance learned through meta-optimization on validation sets will reliably transfer to selecting hard noise samples in the primary training process.

What would settle it

Running the full training pipeline with and without the noise rater on ImageNet and checking if the version using rater-selected noise fails to achieve lower FID scores or slower convergence.

Figures

Figures reproduced from arXiv: 2605.08144 by Aaron Tu, Aditya Joshi, Da Xing, Fang Wu, Hanchen Wang, Hanqun Cao, Haokai Zhao, Hongbin Lin, Jure Leskovec, Kuan Pang, Li Erran Li, Molei Tao, Peng Xia, Tinson Xu, Xiangru Tang, Yanchao Li, Yejin Choi, Yinxi Li, Zehong Wang, Zeqi Zhou.

Figure 1
Figure 1. Figure 1: Illustration of our meta-Learned noise valuation approach for diffusion training. We first [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Rater behavior across diffusion timesteps [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Rater behavior across diffusion training stages, aggregated over all timesteps. (a) At [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The illustration of the noise rater’s architecture, which contains multiple dual-stream DiT [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
read the original abstract

Diffusion models have achieved remarkable success across a wide range of generative tasks, yet their training paradigm largely treats injected noise as uniformly informative. In this work, we challenge this assumption and introduce NoiseRater, a meta-learning framework for instance-level noise valuation in diffusion model training. We propose a parametric noise rater that assigns importance scores to individual noise realizations conditioned on data and timestep, enabling adaptive reweighting of the training objective. The rater is trained via bilevel optimization to improve downstream validation performance after inner-loop diffusion updates. To enable efficient deployment, we further design a decoupled two-stage pipeline that transitions from soft weighting during meta-training to hard noise selection during standard training. Extensive experiments on FFHQ and ImageNet demonstrate that not all noise samples contribute equally, and that prioritizing informative noise improves both training efficiency and generation quality. Our results establish noise valuation as a complementary and previously underexplored axis for improving diffusion model training. Our code is available at: https://anonymous.4open.science/r/NoiseRater-DEB116.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces NoiseRater, a meta-learning framework for instance-level noise valuation in diffusion model training. It trains a parametric noise rater via bilevel optimization to assign importance scores conditioned on data and timestep, enabling adaptive reweighting of the training objective. A decoupled two-stage pipeline transitions from soft weighting during meta-training to hard noise selection during standard training. Experiments on FFHQ and ImageNet are claimed to show that not all noise samples contribute equally and that prioritizing informative noise improves training efficiency and generation quality.

Significance. If the results hold, the work identifies noise valuation as a complementary axis for diffusion model optimization beyond standard uniform noise assumptions, with potential gains in efficiency and FID. The public code link supports reproducibility.

major comments (2)
  1. [Decoupled two-stage pipeline description] The central claim depends on the transfer from soft-reweighted bilevel meta-training (optimized for validation performance) to hard noise selection in deployment, yet no ablation or direct comparison is provided showing that the learned importance scores remain beneficial under the hard-selection regime actually used in standard training; this mismatch in induced training distributions is load-bearing for the efficiency and quality claims.
  2. [Experiments] The abstract asserts positive results on FFHQ and ImageNet, but the manuscript supplies no quantitative tables, ablation details on the bilevel optimization or soft-to-hard transition, statistical tests, or explicit baseline comparisons to uniform noise training, preventing verification of the magnitude and reliability of the reported gains.
minor comments (2)
  1. [Abstract] The code availability link uses an anonymous service; replace with a permanent repository upon acceptance.
  2. [Method] Clarify the exact form of the inner-loop diffusion updates and how the meta-objective is computed in the bilevel setup to avoid ambiguity in the optimization procedure.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback on the NoiseRater manuscript. We address each major comment point by point below, with revisions incorporated to strengthen the presentation of the two-stage pipeline and experimental results.

read point-by-point responses

  1. Referee: [Decoupled two-stage pipeline description] The central claim depends on the transfer from soft-reweighted bilevel meta-training (optimized for validation performance) to hard noise selection in deployment, yet no ablation or direct comparison is provided showing that the learned importance scores remain beneficial under the hard-selection regime actually used in standard training; this mismatch in induced training distributions is load-bearing for the efficiency and quality claims.

    Authors: We agree that an explicit demonstration of the learned scores' benefit under the hard-selection regime is necessary to support the claims. The bilevel optimization is designed to produce scores that improve validation performance after inner-loop updates, providing a principled basis for transfer, but we acknowledge the need for direct evidence. In the revised manuscript, we have added an ablation comparing hard noise selection using the meta-learned NoiseRater scores against uniform random selection and other heuristics during standard training on FFHQ and ImageNet. The results confirm gains in training efficiency and generation quality, validating the soft-to-hard transition. revision: yes

  2. Referee: [Experiments] The abstract asserts positive results on FFHQ and ImageNet, but the manuscript supplies no quantitative tables, ablation details on the bilevel optimization or soft-to-hard transition, statistical tests, or explicit baseline comparisons to uniform noise training, preventing verification of the magnitude and reliability of the reported gains.

    Authors: We have revised the experimental section to include comprehensive quantitative tables reporting FID scores, training efficiency metrics, and generation quality improvements on both FFHQ and ImageNet, with explicit comparisons to uniform noise baselines. Additional ablations detail the bilevel optimization hyperparameters and the soft-to-hard transition effects. Statistical reliability is now shown via means and standard deviations over multiple independent runs. These additions enable full verification of the reported gains. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper's core contribution is a bilevel meta-learning procedure that trains a parametric noise rater on validation performance after inner-loop diffusion updates, followed by an empirical switch to hard selection at deployment. This structure is standard bilevel optimization with held-out validation; no equations reduce the reported efficiency or FID gains to a fitted quantity by construction, nor does any step invoke self-citations, uniqueness theorems, or ansatzes that collapse the claim to its inputs. The derivation remains self-contained against external benchmarks because the meta-objective and final performance metrics are measured on separate data splits and evaluated via standard diffusion training protocols.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on a new parametric noise rater whose weights are fitted during meta-training and on the assumption that bilevel optimization will produce a rater that remains useful after the pipeline is switched to hard selection.

free parameters (1)
  • noise rater parameters
    Weights of the parametric noise rater are learned via the outer loop of bilevel optimization.
axioms (1)
  • domain assumption Bilevel optimization can be solved to produce a noise rater that improves downstream diffusion validation performance.
    The training procedure assumes the outer-loop objective yields a useful rater without specifying convergence guarantees or regularization.

pith-pipeline@v0.9.0 · 5556 in / 1293 out tokens · 84800 ms · 2026-05-12T01:39:52.762738+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

54 extracted references · 54 canonical work pages · 4 internal anchors

  1. [1]

    D. Ahn, J. Kang, S. Lee, J. Min, M. Kim, W. Jang, H. Cho, S. Paul, S. Kim, E. Cha, et al. A noise is worth diffusion guidance.arXiv preprint arXiv:2412.03895, 2024

    work page arXiv 2024

  2. [2]

    Bansal, H.-M

    A. Bansal, H.-M. Chu, A. Schwarzschild, S. Sengupta, M. Goldblum, J. Geiping, and T. Gold- stein. Universal guidance for diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 843–852, 2023

    work page 2023

  3. [3]

    Bechtle, A

    S. Bechtle, A. Molchanov, Y . Chebotar, E. Grefenstette, L. Righetti, G. Sukhatme, and F. Meier. Meta learning via learned loss. In2020 25th International Conference on Pattern Recognition (ICPR), pages 4161–4168. IEEE, 2021

    work page 2021

  4. [4]

    D. A. Calian, G. Farquhar, I. Kemaev, L. M. Zintgraf, M. Hessel, J. Shar, J. Oh, A. György, T. Schaul, J. Dean, et al. Datarater: Meta-learned dataset curation.arXiv preprint arXiv:2505.17895, 2025

    work page arXiv 2025

  5. [5]

    C. Chen, L. Yang, X. Yang, L. Chen, G. He, C. Wang, and Y . Li. Find: Fine-tuning initial noise distribution with policy optimization for diffusion models. InProceedings of the 32nd ACM International Conference on Multimedia, pages 6735–6744, 2024

    work page 2024

  6. [6]

    J. Choi, J. Lee, C. Shin, S. Kim, H. Kim, and S. Yoon. Perception prioritized training of diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11472–11481, 2022

    work page 2022

  7. [7]

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei. Imagenet: A large-scale hierarchical image database. In2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009

    work page 2009

  8. [8]

    Elata, T

    N. Elata, T. Michaeli, and M. Elad. Adaptive compressed sensing with diffusion-based posterior sampling. InEuropean Conference on Computer Vision, pages 290–308. Springer, 2024

    work page 2024

  9. [9]

    Engstrom, A

    L. Engstrom, A. Ilyas, B. Chen, A. Feldmann, W. Moses, and A. Madry. Optimizing ml training with metagradient descent.arXiv preprint arXiv:2503.13751, 2025

    work page arXiv 2025

  10. [10]

    Esser, S

    P. Esser, S. Kulal, A. Blattmann, R. Entezari, J. Müller, H. Saini, Y . Levi, D. Lorenz, A. Sauer, F. Boesel, et al. Scaling rectified flow transformers for high-resolution image synthesis. In Forty-first international conference on machine learning, 2024

    work page 2024

  11. [11]

    Eyring, V

    L. Eyring, V . Pauline, S. Bauer, Z. Akata, and A. Dosovitskiy. Ddno: Discrete diffusion noise optimization

    work page

  12. [12]

    X. Guo, J. Liu, M. Cui, J. Li, H. Yang, and D. Huang. Initno: Boosting text-to-image diffusion models via initial noise optimization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9380–9389, 2024

    work page 2024

  13. [13]

    T. Hang, S. Gu, C. Li, J. Bao, D. Chen, H. Hu, X. Geng, and B. Guo. Efficient diffusion training via min-snr weighting strategy. InProceedings of the IEEE/CVF international conference on computer vision, pages 7441–7451, 2023

    work page 2023

  14. [14]

    T. Hang, S. Gu, J. Bao, F. Wei, D. Chen, X. Geng, and B. Guo. Improved noise schedule for diffusion training. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 4796–4806, 2025

    work page 2025

  15. [15]

    H. He, J. Liang, X. Wang, P. Wan, D. Zhang, K. Gai, and L. Pan. Scaling image and video generation via test-time evolutionary search.arXiv preprint arXiv:2505.17618, 2025. 11

    work page arXiv 2025

  16. [16]

    Heusel, H

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium.Advances in neural information processing systems, 30, 2017

    work page 2017

  17. [17]

    Classifier-Free Diffusion Guidance

    J. Ho and T. Salimans. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  18. [18]

    J. Ho, A. Jain, and P. Abbeel. Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    work page 2020

  19. [19]

    Jiang, Z

    L. Jiang, Z. Zhou, T. Leung, L.-J. Li, and L. Fei-Fei. Mentornet: Learning data-driven curriculum for very deep neural networks on corrupted labels. InInternational conference on machine learning, pages 2304–2313. PMLR, 2018

    work page 2018

  20. [20]

    Karras, S

    T. Karras, S. Laine, and T. Aila. A style-based generator architecture for generative adversar- ial networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4401–4410, 2019

    work page 2019

  21. [21]

    Karras, M

    T. Karras, M. Aittala, T. Aila, and S. Laine. Elucidating the design space of diffusion-based generative models.Advances in neural information processing systems, 35:26565–26577, 2022

    work page 2022

  22. [22]

    Karunratanakul, K

    K. Karunratanakul, K. Preechakul, E. Aksan, T. Beeler, S. Suwajanakorn, and S. Tang. Opti- mizing diffusion noise can serve as universal motion priors. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1334–1345, 2024

    work page 2024

  23. [23]

    Kemaev, D

    I. Kemaev, D. A. Calian, L. M. Zintgraf, G. Farquhar, and H. Van Hasselt. Scalable meta-learning via mixed-mode differentiation.arXiv preprint arXiv:2505.00793, 2025

    work page arXiv 2025

  24. [24]

    J. Kim, T. Yoon, J. Hwang, and M. Sung. Inference-time scaling for flow models via stochastic generation and rollover budget forcing.arXiv preprint arXiv:2503.19385, 2025

    work page arXiv 2025

  25. [25]

    J.-Y . Kim, H. Go, S. Kwon, and H.-G. Kim. Denoising task difficulty-based curriculum for training diffusion models.arXiv preprint arXiv:2403.10348, 2024

    work page arXiv 2024

  26. [26]

    X. Li, M. Uehara, X. Su, G. Scalia, T. Biancalani, A. Regev, S. Levine, and S. Ji. Dynamic search for inference-time alignment in diffusion models.arXiv preprint arXiv:2503.02039, 2025

    work page arXiv 2025

  27. [27]

    Y . Li, H. Jiang, A. Kodaira, M. Tomizuka, K. Keutzer, and C. Xu. Immiscible diffusion: Accel- erating diffusion training with noise assignment.Advances in neural information processing systems, 37:90198–90225, 2024

    work page 2024

  28. [28]

    Y . Li, F. Liang, D. Kondratyuk, M. Tomizuka, K. Keutzer, and C. Xu. Improved immiscible diffu- sion: Accelerate diffusion training by reducing its miscibility.arXiv preprint arXiv:2505.18521, 2025

    work page arXiv 2025

  29. [29]

    N. Ma, S. Tong, H. Jia, H. Hu, Y .-C. Su, M. Zhang, X. Yang, Y . Li, T. Jaakkola, X. Jia, et al. Scaling inference time compute for diffusion models. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 2523–2534, 2025

    work page 2025

  30. [30]

    Maclaurin, D

    D. Maclaurin, D. Duvenaud, and R. Adams. Gradient-based hyperparameter optimization through reversible learning. InInternational conference on machine learning, pages 2113–2122. PMLR, 2015

    work page 2015

  31. [31]

    Peebles and S

    W. Peebles and S. Xie. Scalable diffusion models with transformers. InProceedings of the IEEE/CVF international conference on computer vision, pages 4195–4205, 2023

    work page 2023

  32. [32]

    Movie Gen: A Cast of Media Foundation Models

    A. Polyak, A. Zohar, A. Brown, A. Tjandra, A. Sinha, A. Lee, A. Vyas, B. Shi, C.-Y . Ma, C.-Y . Chuang, et al. Movie gen: A cast of media foundation models.arXiv preprint arXiv:2410.13720, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  33. [33]

    Z. Qi, L. Bai, H. Xiong, and Z. Xie. Not all noises are created equally: Diffusion noise selection and optimization.arXiv preprint arXiv:2407.14041, 2024. 12

    work page arXiv 2024

  34. [34]

    Ramesh, M

    A. Ramesh, M. Pavlov, G. Goh, S. Gray, C. V oss, A. Radford, M. Chen, and I. Sutskever. Zero-shot text-to-image generation. InInternational conference on machine learning, pages 8821–8831. Pmlr, 2021

    work page 2021

  35. [35]

    G. Raya, B. Nguyen, G. Batzolis, Y . Takida, D. Stancevic, N. Murata, C.-H. Lai, Y . Mitsufuji, and L. Ambrogioni. Information-guided noise allocation for efficient diffusion training.arXiv preprint arXiv:2602.18647, 2026

    work page arXiv 2026

  36. [36]

    M. Ren, W. Zeng, B. Yang, and R. Urtasun. Learning to reweight examples for robust deep learning. InInternational conference on machine learning, pages 4334–4343. PMLR, 2018

    work page 2018

  37. [37]

    Y . Ren, W. Gao, L. Ying, G. M. Rotskoff, and J. Han. Driftlite: Lightweight drift control for inference-time scaling of diffusion models.arXiv preprint arXiv:2509.21655, 2025

    work page arXiv 2025

  38. [38]

    Rojas, Y

    K. Rojas, Y . Zhu, S. Zhu, F. X.-F. Ye, and M. Tao. Diffuse everything: Multimodal diffusion models on arbitrary state spaces.ICML, 2025

    work page 2025

  39. [39]

    Rombach, A

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

    work page 2022

  40. [40]

    J. Shu, Q. Xie, L. Yi, Q. Zhao, S. Zhou, Z. Xu, and D. Meng. Meta-weight-net: Learning an explicit mapping for sample weighting.Advances in neural information processing systems, 32, 2019

    work page 2019

  41. [41]

    arXiv preprint arXiv:2501.06848 (2025)

    R. Singhal, Z. Horvitz, R. Teehan, M. Ren, Z. Yu, K. McKeown, and R. Ranganath. A general framework for inference-time scaling and steering of diffusion models.arXiv preprint arXiv:2501.06848, 2025

    work page arXiv 2025

  42. [42]

    Sohl-Dickstein, E

    J. Sohl-Dickstein, E. Weiss, N. Maheswaranathan, and S. Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. InInternational conference on machine learning, pages 2256–2265. pmlr, 2015

    work page 2015

  43. [43]

    J. Song, C. Meng, and S. Ermon. Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  44. [44]

    Y . Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole. Score-based generative modeling through stochastic differential equations.arXiv preprint arXiv:2011.13456, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  45. [45]

    Stecklov, N

    A. Stecklov, N. E. Rimawi-Fine, and M. Blanchette. Inference-time compute scaling for flow matching.arXiv preprint arXiv:2510.17786, 2025

    work page arXiv 2025

  46. [46]

    Sun and L

    N. Sun and L. Shi. Variance-aware adaptive weighting for diffusion model training.arXiv preprint arXiv:2603.10391, 2026

    work page arXiv 2026

  47. [47]

    Q. Sun, Z. Jiang, H. Zhao, and K. He. Is noise conditioning necessary for denoising generative models?arXiv preprint arXiv:2502.13129, 2025

    work page arXiv 2025

  48. [48]

    Z. Tang, J. Peng, J. Tang, M. Hong, F. Wang, and T.-H. Chang. Tuning-free alignment of diffusion models with direct noise optimization. InICML 2024 Workshop on Structured Probabilistic Inference{\&}Generative Modeling, 2024

    work page 2024

  49. [49]

    Y . Wang, Y . He, and M. Tao. Evaluating the design space of diffusion-based generative models. Advances in Neural Information Processing Systems, 37:19307–19352, 2024

    work page 2024

  50. [50]

    Z. Wang, G. Hu, and Q. Hu. Training noise-robust deep neural networks via meta-learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4524–4533, 2020

    work page 2020

  51. [51]

    J. L. Watson, D. Juergens, N. R. Bennett, B. L. Trippe, J. Yim, H. E. Eisenach, W. Ahern, A. J. Borst, R. J. Ragotte, L. F. Milles, et al. De novo design of protein structure and function with rfdiffusion.Nature, 620(7976):1089–1100, 2023. 13

    work page 2023

  52. [52]

    L. Yang, Y . Tian, B. Li, X. Zhang, K. Shen, Y . Tong, and M. Wang. Mmada: Multimodal large diffusion language models.arXiv preprint arXiv:2505.15809, 2025

    work page arXiv 2025

  53. [53]

    J. Yoon, H. Cho, D. Baek, Y . Bengio, and S. Ahn. Monte carlo tree diffusion for system 2 planning.arXiv preprint arXiv:2502.07202, 2025

    work page arXiv 2025

  54. [54]

    Z. Zhou, S. Shao, L. Bai, S. Zhang, Z. Xu, B. Han, and Z. Xie. Golden noise for diffusion models: A learning framework. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 17688–17697, 2025. 14 A Mathematical Analysis A.1 Proof of Thm. 4.1 Define F(θ, η) :=∇ θLinner(θ;η) . At the inner optimum θ∗(η), the first-order optimality...

    work page 2025