Pith. sign in

REVIEW 4 major objections 5 minor 54 references

Remix-DiT: Mixing Diffusion Transformers for Multi-Expert Denoising

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read Remix-DiT argues that four shared basis transformers, mixed by learned coefficients into twenty timestep experts, improve ImageNet image quality at the same per-step inference cost as a single diffusion transformer.

desk verdict Remix-DiT is a clean, honest empirical paper that shows real FID gains by mixing a few shared DiT bases into many timestep experts, though the size of the improvement over the best-controlled baseline is modest and single-run results leave some uncertainty. read the letter →

arxiv 2412.05628 v1 pith:A6CLN7ZI submitted 2024-12-07 cs.CV cs.AI

classification cs.CVcs.AI
keywords multi-expertdenoisingdiffusiontransformersparametermixingmodelmergingtimestepexpertslearnablecoefficientsImageNetgenerationfine-tuningmodels
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 sets out to show that a diffusion transformer does not have to be a single monolithic denoiser to benefit from multi-expert denoising. Its proposal, Remix-DiT, trains K shared basis transformers and a small learned table of softmax mixing coefficients, then crafts N timestep experts by taking weighted averages of the K basis parameter sets ($\Theta_{N\times P} = \alpha_{N\times K}\beta_{K\times P}$). The central claim is that this learned allocation of capacity across timesteps improves generation quality over both the original transformer and independently trained experts, while keeping inference exactly as cheap as a standard transformer because each mixed expert has the same architecture. If true, the practical payoff is that a practitioner can choose a generous number of experts without paying N times the training cost, and the coefficients, rather than a human schedule, decide which timesteps get specialized versus blended capacity. The paper backs the claim with 100K-step fine-tunes of pretrained DiT-S/B/L models on ImageNet 256 by 256, reporting lower FID than continual training and than equal-budget multi-expert baselines.

What carries the argument

The load-bearing object is the mixing identity $\Theta_{N\times P}=\alpha_{N\times K}\beta_{K\times P}$: an $N\times K$ matrix of softmax-normalized logits, implemented as an embedding layer, multiplied by K flattened basis parameter sets. This lets the number of experts grow without growing trained parameters. It is paired with a construction trick that packs the K bases into one DiT whose linear layers are K times wider, so a forward pass first performs the weighted average and then runs a standard-width linear layer; once experts are precomputed, per-step inference cost matches a plain DiT. A one-hot prior on the coefficients, annealed away during training, forces the K bases to separate early instead of stalling as near-identical copies of the pretrained checkpoint, and hierarchical sampling of an expert interval followed by a timestep within it lets all bases receive gradients at every step.

What would settle it

Train K=4 bases from the same initialization on a domain where low- and high-noise denoising require qualitatively different operations, such as text or audio, then check whether the mixed experts can match per-interval loss of experts trained independently on single intervals; any interval where the mixture is systematically worse would falsify the simplex-span assumption.

Watch

Extended reading notes

Core claim

The core discovery is that expert denoisers for different noise intervals can be manufactured rather than trained. Writing every expert parameter vector as a linear mixture $\theta_i = \sum_k \alpha_{ik}\beta_k$ of K basis vectors, and learning the coefficients together with the bases, produces experts that specialize by interval while sharing all parameters with a single K-times-wider DiT. On ImageNet 256 by 256, a 100K-step fine-tune with K=4 and N=20 improves FID from 10.11 to 9.02 for DiT-B/2 and from 26.51 to 22.84 for DiT-S/2 relative to the pretrained checkpoints. The learned coefficients concentrate near one-hot at low-noise timesteps and blend bases at high-noise timesteps, and each mixed expert attains its lowest prediction loss inside its own interval, indicating specialization rather than a generic average. Because redundant intervals learn similar coefficients, the method also removes the need to guess the optimal partition of timesteps in advance.

Load-bearing premise

The load-bearing premise is that every useful timestep specialist can be written as a weighted average of a small number K of shared basis networks, and that four bases starting from identical pretrained weights can diverge enough in 100K fine-tuning steps to cover those specialists.

Editorial extensions

If this is right

  • A pretrained diffusion transformer can be improved by a 100K-step fine-tune that mixes four copies of its own weights, without changing the architecture used at inference.
  • The number of experts becomes a free design choice: with K=4, increasing N from 4 to 20 improves FID, while N=1000 hurts, so the limit on quality is the span of the bases rather than the number of experts.
  • Because only one expert is active per step and precomputed experts have the same per-step cost as the original model, the inference budget of the improved model matches the standard DiT.
  • The learned coefficient pattern, one-hot at low-noise steps and blended at high-noise steps, identifies which stages of denoising need dedicated capacity and offers a learned alternative to hand-designed interval partitions.

Reading between the lines

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

  • Beyond the paper, the learned coefficient table reads like a schedule of denoising difficulty and might transfer across datasets or checkpoints without retraining.
  • Beyond the paper, the reported N=1000 degradation suggests a testable scaling law: at fixed K, quality should plateau and then fall as N grows, because the bottleneck is the span of K bases, not the resolution of the timestep partition.
  • Beyond the paper, layer-wise coefficient routing remains an open avenue: the paper's local mixer underperformed its global mixer in a short 10K-step comparison, but with longer training the extra expressiveness of per-layer mixing could reverse that ordering.
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 Remix-DiT, a method to construct N denoising experts for different timestep intervals from K (K<N) basis DiT models via learnable mixing coefficients. The experts are parameter-weighted averages of the basis models; a softmax over mixing logits ensures convexity, and a one-hot prior with linear annealing encourages basis specialization. The models are initialized from a pretrained DiT by replicating weights K times and then fine-tuned for 100K steps. Experiments on ImageNet 256x256 with DiT-S/B/L report FID/IS gains over the base DiT, continual training, and an 8-expert independently trained multi-expert baseline. Ablations address mixer type, global vs local coefficients, and choices of N and K, with an efficiency table comparing training/inference latency and memory.

Significance. If the results hold, the method offers a practical way to obtain part of the benefit of multi-expert denoising while training only K shared bases and keeping inference FLOPs comparable to a single DiT. The paper ships code, spans three model sizes, and includes systematic ablations over mixer type, N, and K, which are useful for the community. The learned coefficient patterns provide qualitative evidence that the method discovers a soft timestep specialization. The main risk is that the central comparison to independently trained experts is confounded by the number of gradient updates each parameter receives, and the expressiveness of the K-basis convex hull is never directly tested.

major comments (4)
  1. [§5.2, Table 1] The comparison to the 'Multi Experts' baseline is confounded. In Table 1, the 8 independent experts are each trained for 12.5K steps (100K total), whereas each of the 4 Remix bases receives 100K gradient-updating steps because every basis is updated at every training step. The reported FID advantage of Remix (e.g., 9.02 vs 9.28 for DiT-B) could therefore be caused by the larger number of updates per parameter set rather than by the mixing parameterization. The paper should add a matched-budget baseline in which independent experts are trained for the same total compute with equal per-expert steps (e.g., 8 experts × 100K steps or 4 experts × 100K steps), as well as a Remix variant with N=8 to separate the effect of the number of experts from the mixing mechanism.
  2. [§5.4, Table 2] The Onehot Mixer ablation does not control for per-basis update counts. With N=20 and K=4 under uniform interval sampling, each one-hot basis is updated only about 2.5K of the 10K fine-tuning steps, whereas every softmax basis is updated at every step. The FID difference between Onehot (32.00) and Softmax (31.11) may therefore reflect the number of gradient updates each basis receives, not the expressiveness of softmax mixing. To support the claim that the mixing mechanism itself is responsible, the authors should report a one-hot mixer with matched per-basis updates (e.g., an independent 4-expert baseline trained for 10K steps per expert, or a one-hot setup with interval sampling proportional to basis assignment).
  3. [§4, Eq. (4)] The central claim that Remix-DiT 'achieves superior generation results compared to independently trained multiple expert models' rests on the assumption that useful timestep experts can be represented as convex combinations of K=4 basis parameter vectors. The paper offers no direct evidence for this representational assumption: it never measures how well the convex hull of learned bases can approximate independently trained experts, and the non-monotonic behavior in Table 2 (Remix-S-2-20: 31.67, Remix-S-4-20: 31.11, Remix-S-8-20: 32.05) is hard to reconcile with a pure expressiveness story. A direct test, such as fitting independently pretrained experts with the mixing model and reporting approximation error, or comparing against an unconstrained low-rank factorization of expert parameters, would substantiate the claim and clarify the role of K.
  4. [§5 and NeurIPS checklist answer 7] All FID/IS/Precision/Recall numbers are single-run measurements, and the authors explicitly answer 'No' to the statistical-significance question. Many of the reported margins are small (e.g., FID 9.02 vs 9.28 in Table 1, and 31.11 vs 31.21 in the Global/Local ablation). Without error bars or a significance test, the robustness of these improvements is unassessed. The authors should provide at least three seeds with standard deviations for the main tables, or a paired test across samples, to establish that the observed gains are not noise.
minor comments (5)
  1. [§4, Eq. (5)] The gradient formula appears to contain a typo: the right-hand side sums over k but the left-hand side has a fixed k, and the element-wise product with β_k should be an inner product for the scalar coefficient gradient. Please correct the equation and the surrounding explanation.
  2. [§5.4, paragraph before Table 2] The text uses 'RemixDiT' and 'Remix-DiT' inconsistently (also 'Remix-B' vs 'Remix-B/2-4-20'). Please unify the notation across the paper.
  3. [Figure 3 caption] The word 'Comparision' should be 'Comparison'.
  4. [§5.1] The paper does not specify the FID computation for Table 1 (number of generated samples, sampling steps, and whether FID-10K or FID-50K is used). Please add this information for reproducibility.
  5. [§5.2 / Table 3] When precomputing N experts for inference, the storage cost is N× the model parameters; Table 3 reports only training GPU memory and per-step latency. Please include the inference storage requirement and clarify the trade-off between runtime mixing and precomputation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Remix-DiT's central claims are supported by external benchmark measurements; self-citations are motivational only.

full rationale

The paper's derivation chain is empirical and self-contained. The central object, Eq. 4, defines each expert as a weighted average of K learned basis parameter sets, with the mixing coefficients trained by minimizing the standard denoising objective (Eqs. 1 and 2) on ImageNet; no coefficient is fit to the evaluation metric (FID/IS). The reported quality numbers are external benchmark measurements comparing against standard DiT, continual training, and multi-expert baselines, so the central claim of improved generation quality does not reduce to the model's own definition. The cited works [36] and [38], including self-citations, motivate parameter fusion and the width-expansion packing trick, but they are not used to derive the empirical gains; those gains are measured directly. The one-hot prior in Eq. 6 is an initialization and a regularization term that is annealed to zero, and the observed 'specialization' of experts is a post-hoc analysis rather than a predicted quantity derived from the prior. The skeptic's concern that four basis models may not span the useful expert manifold, or that the budget comparison favors shared bases, is a representational-capacity and experimental-fairness question, not a circularity. Therefore no circular step is present.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claim rests on two empirical premises: that convex combinations of K DiT parameter sets are useful denoisers for all timestep intervals, and that a shared pretrained initialization can differentiate within 100K steps. The only hand-set hyperparameters are K, N, and the regularization strength gamma; a third free parameter, gamma, is not reported numerically. No new physical or architectural entities are postulated.

free parameters (3)
  • K (number of basis models) = 4
    K=4 used in main results; ablation in Table 2 shows K=4 beats K=2 and K=8 for FID-10K.
  • N (number of experts) = 20
    N=20 used in main experiments; ablation shows N=20 best among 4, 8, 20, 100, 1000.
  • gamma and its annealing schedule = not specified in text
    Eq. 6 adds cross-entropy prior with strength gamma annealed to 0; the exact value and schedule are omitted, so the reader cannot assess its impact.
assumptions (3)
  • standard math DDPM denoising objective (Eq. 1) and timestep-conditional network are taken from prior work [13,29].
    Section 3 restates Eq. 1 without derivation.
  • domain assumption Parameter-space averaging of networks preserves or improves the denoising function.
    Section 4 Eq. 4 relies on this to craft experts; cited [36,38] but not proven here.
  • ad hoc to paper One-hot prior regularization forces bases to specialize, with annealing to gamma=0 keeping diversity.
    Section 5.2 describes this prior; no analysis of sensitivity to gamma or the annealing schedule.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Remix-DiT: Mixing Diffusion Transformers for Multi-Expert Denoising." pith.science (2026). https://pith.science/paper/A6CLN7ZI

@misc{pith2026241205628,
  author       = {Pith},
  title        = {Pith review of: Remix-DiT: Mixing Diffusion Transformers for Multi-Expert Denoising},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A6CLN7ZI}},
  note         = {Machine review of arXiv:2412.05628}
}
read the original abstract

Transformer-based diffusion models have achieved significant advancements across a variety of generative tasks. However, producing high-quality outputs typically necessitates large transformer models, which result in substantial training and inference overhead. In this work, we investigate an alternative approach involving multiple experts for denoising, and introduce Remix-DiT, a novel method designed to enhance output quality at a low cost. The goal of Remix-DiT is to craft N diffusion experts for different denoising timesteps, yet without the need for expensive training of N independent models. To achieve this, Remix-DiT employs K basis models (where K < N) and utilizes learnable mixing coefficients to adaptively craft expert models. This design offers two significant advantages: first, although the total model size is increased, the model produced by the mixing operation shares the same architecture as a plain model, making the overall model as efficient as a standard diffusion transformer. Second, the learnable mixing adaptively allocates model capacity across timesteps, thereby effectively improving generation quality. Experiments conducted on the ImageNet dataset demonstrate that Remix-DiT achieves promising results compared to standard diffusion transformers and other multiple-expert methods. The code is available at https://github.com/VainF/Remix-DiT.

Figures

Figures reproduced from arXiv: 2412.05628 by the authors.

Figure 1
Figure 1. (a) Re-training a larger DiT incurs significant training and inference costs. (b) Multi [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. An example of mixing 4 linear layers basis into 6 expert layers. Each expert linear layer is [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Comparision to existing methods To further investigate the behavior of the experts learned by the algorithm, we visualized the mixing coefficients. As shown in Figure 4a, we conducted experiments on DiT-S. During the training of 20 ex￾pert models, we observed that the algorithm assigned more one-hot coefficients to timesteps close to 0. At these steps, the denoising model focuses more on high-frequency detail featur… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: (a) The learned coefficients for a Remix-DiT-S-4-20, which mixes 4 basis DiT-S to obtain [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Visualization of generated samples from DiT-B and Remix-DiT-B. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 24 canonical work pages

  1. [1]

    ediff-i: Text-to-image diffusion models with an ensemble of expert denoisers

    Yogesh Balaji, Seungjun Nah, Xun Huang, Arash Vahdat, Jiaming Song, Qinsheng Zhang, Karsten Kreis, Miika Aittala, Timo Aila, Samuli Laine, et al. ediff-i: Text-to-image diffusion models with an ensemble of expert denoisers. arXiv preprint arXiv:2211.01324, 2022

  2. [2]

    All are worth words: A vit backbone for diffusion models

    Fan Bao, Shen Nie, Kaiwen Xue, Yue Cao, Chongxuan Li, Hang Su, and Jun Zhu. All are worth words: A vit backbone for diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 22669–22679, 2023

  3. [3]

    Video generation models as world simulators

    Tim Brooks, Bill Peebles, Connor Holmes, Will DePue, Yufei Guo, Li Jing, David Schnurr, Joe Taylor, Troy Luhman, Eric Luhman, Clarence Ng, Ricky Wang, and Aditya Ramesh. Video generation models as world simulators. 2024

  4. [4]

    Exploring Vision Transformers as Diffusion Learners

    He Cao, Jianan Wang, Tianhe Ren, Xianbiao Qi, Yihao Chen, Yuan Yao, and Lei Zhang. Exploring vision transformers as diffusion learners. arXiv preprint arXiv:2212.13771, 2022

  5. [5]

    Gentron: Delving deep into diffusion transformers for image and video generation

    Shoufa Chen, Mengmeng Xu, Jiawei Ren, Yuren Cong, Sen He, Yanping Xie, Animesh Sinha, Ping Luo, Tao Xiang, and Juan-Manuel Perez-Rua. Gentron: Delving deep into diffusion transformers for image and video generation. arXiv preprint arXiv:2312.04557, 2023

  6. [6]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009

  7. [7]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. Advances in neural information processing systems, 34:8780–8794, 2021

  8. [8]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020

Show all 54 references
  1. [9]

    Structural pruning for diffusion models

    Gongfan Fang, Xinyin Ma, and Xinchao Wang. Structural pruning for diffusion models. In Advances in Neural Information Processing Systems, 2023

  2. [10]

    Masked diffusion transformer is a strong image synthesizer

    Shanghua Gao, Pan Zhou, Ming-Ming Cheng, and Shuicheng Yan. Masked diffusion transformer is a strong image synthesizer. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 23164–23173, 2023

  3. [11]

    Vector quantized diffusion model for text-to-image synthesis

    Shuyang Gu, Dong Chen, Jianmin Bao, Fang Wen, Bo Zhang, Dongdong Chen, Lu Yuan, and Baining Guo. Vector quantized diffusion model for text-to-image synthesis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10696–10706, 2022

  4. [12]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick. Masked autoencoders are scalable vision learners. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16000–16009, 2022

  5. [14]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020

  6. [15]

    U-dit tts: U-diffusion vision transformer for text-to-speech

    Xin Jing, Yi Chang, Zijiang Yang, Jiangjian Xie, Andreas Triantafyllopoulos, and Bjoern W Schuller. U-dit tts: U-diffusion vision transformer for text-to-speech. In Speech Communication; 15th ITG Conference, pages 56–60. VDE, 2023

  7. [16]

    Multi- architecture multi-expert diffusion models

    Yunsung Lee, JinYoung Kim, Hyojun Go, Myeongho Jeong, Shinhyeok Oh, and Seungtaek Choi. Multi- architecture multi-expert diffusion models. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 13427–13436, 2024

  8. [17]

    Dlt: Conditioned layout generation with joint discrete-continuous diffusion layout transformer

    Elad Levi, Eli Brosh, Mykola Mykhailych, and Meir Perez. Dlt: Conditioned layout generation with joint discrete-continuous diffusion layout transformer. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 2106–2115, 2023

  9. [18]

    Oms-dpm: Optimizing the model schedule for diffusion probabilistic models

    Enshu Liu, Xuefei Ning, Zinan Lin, Huazhong Yang, and Yu Wang. Oms-dpm: Optimizing the model schedule for diffusion probabilistic models. arXiv preprint arXiv:2306.08860, 2023

  10. [19]

    Vit-tts: visual text-to-speech with scalable diffusion transformer

    Huadai Liu, Rongjie Huang, Xuan Lin, Wenqiang Xu, Maozong Zheng, Hong Chen, Jinzheng He, and Zhou Zhao. Vit-tts: visual text-to-speech with scalable diffusion transformer. arXiv preprint arXiv:2305.12708, 2023

  11. [20]

    Vdt: General- purpose video diffusion transformers via mask modeling

    Haoyu Lu, Guoxing Yang, Nanyi Fei, Yuqi Huo, Zhiwu Lu, Ping Luo, and Mingyu Ding. Vdt: General- purpose video diffusion transformers via mask modeling. In The Twelfth International Conference on Learning Representations, 2023

  12. [21]

    Latte: Latent diffusion transformer for video generation

    Xin Ma, Yaohui Wang, Gengyun Jia, Xinyuan Chen, Ziwei Liu, Yuan-Fang Li, Cunjian Chen, and Yu Qiao. Latte: Latent diffusion transformer for video generation. arXiv preprint arXiv:2401.03048, 2024. 11

  13. [22]

    Deepcache: Accelerating diffusion models for free

    Xinyin Ma, Gongfan Fang, and Xinchao Wang. Deepcache: Accelerating diffusion models for free. arXiv preprint arXiv:2312.00858, 2023

  14. [23]

    Deepcache: Accelerating diffusion models for free

    Xinyin Ma, Gongfan Fang, and Xinchao Wang. Deepcache: Accelerating diffusion models for free. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15762– 15772, 2024

  15. [24]

    Snap video: Scaled spatiotemporal transformers for text-to-video synthesis

    Willi Menapace, Aliaksandr Siarohin, Ivan Skorokhodov, Ekaterina Deyneka, Tsai-Shien Chen, Anil Kag, Yuwei Fang, Aleksei Stoliar, Elisa Ricci, Jian Ren, et al. Snap video: Scaled spatiotemporal transformers for text-to-video synthesis. arXiv preprint arXiv:2402.14797, 2024

  16. [25]

    Dit-3d: Exploring plain diffusion transformers for 3d shape generation.Advances in Neural Information Processing Systems, 36, 2024

    Shentong Mo, Enze Xie, Ruihang Chu, Lanqing Hong, Matthias Niessner, and Zhenguo Li. Dit-3d: Exploring plain diffusion transformers for 3d shape generation.Advances in Neural Information Processing Systems, 36, 2024

  17. [26]

    Improved denoising diffusion probabilistic models

    Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion probabilistic models. In International conference on machine learning, pages 8162–8171. PMLR, 2021

  18. [27]

    T-stitch: Accelerating sampling in pre-trained diffusion models with trajectory stitching

    Zizheng Pan, Bohan Zhuang, De-An Huang, Weili Nie, Zhiding Yu, Chaowei Xiao, Jianfei Cai, and Anima Anandkumar. T-stitch: Accelerating sampling in pre-trained diffusion models with trajectory stitching. arXiv preprint arXiv:2402.14167, 2024

  19. [28]

    Denoising task routing for diffusion models

    Byeongjun Park, Sangmin Woo, Hyojun Go, Jin-Young Kim, and Changick Kim. Denoising task routing for diffusion models. arXiv preprint arXiv:2310.07138, 2023

  20. [29]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4195–4205, 2023

  21. [30]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

  22. [31]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part ...

  23. [32]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502, 2020

  24. [33]

    Generative modeling by estimating gradients of the data distribution

    Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. Advances in neural information processing systems, 32, 2019

  25. [34]

    Score-based generative modeling through stochastic differential equations

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. arXiv preprint arXiv:2011.13456, 2020

  26. [35]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in neural information processing systems, 30, 2017

  27. [36]

    Factorizing knowledge in neural networks

    Xingyi Yang, Jingwen Ye, and Xinchao Wang. Factorizing knowledge in neural networks. InEuropean Conference on Computer Vision, pages 73–91. Springer, 2022

  28. [37]

    Diffusion probabilistic model made slim

    Xingyi Yang, Daquan Zhou, Jiashi Feng, and Xinchao Wang. Diffusion probabilistic model made slim. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 22552– 22562, 2023

  29. [38]

    Deep model reassembly

    Xingyi Yang, Daquan Zhou, Songhua Liu, Jingwen Ye, and Xinchao Wang. Deep model reassembly. Advances in neural information processing systems, 35:25739–25753, 2022

  30. [39]

    Your vit is secretly a hybrid discriminative-generative diffusion model

    Xiulong Yang, Sheng-Min Shih, Yinlin Fu, Xiaoting Zhao, and Shihao Ji. Your vit is secretly a hybrid discriminative-generative diffusion model. arXiv preprint arXiv:2208.07791, 2022

  31. [40]

    Fast training of diffusion models with masked transformers

    Hongkai Zheng, Weili Nie, Arash Vahdat, and Anima Anandkumar. Fast training of diffusion models with masked transformers. arXiv preprint arXiv:2306.09305, 2023. 12

  32. [41]

    Guidelines: • The answer NA means that the abstract and introduction do not include the claims made in the paper

    Claims Question: Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? Answer: [Yes] Justification: This submission introduced a new method for multi-expert denoising. Guidelines: • The answer NA means that the abstrac...

  33. [42]

    Limitations

    Limitations Question: Does the paper discuss the limitations of the work performed by the authors? Answer: [Yes] Justification: A limitation section is included. Guidelines: • The answer NA means that the paper has no limitation while the answer No means that the paper has lim...

  34. [43]

    Guidelines: 13 • The answer NA means that the paper does not include theoretical results

    Theory Assumptions and Proofs Question: For each theoretical result, does the paper provide the full set of assumptions and a complete (and correct) proof? Answer: [NA] Justification: This work does not involve theoretical results. Guidelines: 13 • The answer NA means that the...

  35. [44]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experimental Result Reproducibility Question: Does the paper fully disclose all the information needed to reproduce the main ex- perimental results of the paper to the extent that it affects the main claims and/or conclusions of the paper (regardless of whether the code and da...

  36. [45]

    14 Guidelines: • The answer NA means that paper does not include experiments requiring code

    Open access to data and code Question: Does the paper provide open access to the data and code, with sufficient instruc- tions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: Code is available in the su...

  37. [46]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experimental Setting/Details Question: Does the paper specify all the training and test details (e.g., data splits, hyper- parameters, how they were chosen, type of optimizer, etc.) necessary to understand the results? Answer: [Yes] Justification: Training details is summarize...

  38. [47]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experiment Statistical Significance Question: Does the paper report error bars suitably and correctly defined or other appropriate information about the statistical significance of the experiments? Answer: [No] Justification: This submission does not include error bars. Guidel...

  39. [48]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experiments Compute Resources Question: For each experiment, does the paper provide sufficient information on the com- puter resources (type of compute workers, memory, time of execution) needed to reproduce the experiments? Answer: [Yes] Justification: The number of GPUs, tra...

  40. [49]

    Guidelines: • The answer NA means that the authors have not reviewed the NeurIPS Code of Ethics

    Code Of Ethics Question: Does the research conducted in the paper conform, in every respect, with the NeurIPS Code of Ethics https://neurips.cc/public/EthicsGuidelines? Answer: [Yes] Justification: The research was conducted with the NeurIPS Code of Ethics. Guidelines: • The a...

  41. [50]

    Guidelines: • The answer NA means that there is no societal impact of the work performed

    Broader Impacts Question: Does the paper discuss both potential positive societal impacts and negative societal impacts of the work performed? Answer: [Yes] Justification: We discuss the societal impacts in the main paper. Guidelines: • The answer NA means that there is no soc...

  42. [51]

    Guidelines: • The answer NA means that the paper poses no such risks

    Safeguards Question: Does the paper describe safeguards that have been put in place for responsible release of data or models that have a high risk for misuse (e.g., pretrained language models, image generators, or scraped datasets)? Answer: [NA] Justification: the paper poses...

  43. [52]

    Guidelines: • The answer NA means that the paper does not use existing assets

    Licenses for existing assets Question: Are the creators or original owners of assets (e.g., code, data, models), used in the paper, properly credited and are the license and terms of use explicitly mentioned and properly respected? Answer: [Yes] Justification: All papers and a...

  44. [53]

    17 Guidelines: • The answer NA means that the paper does not release new assets

    New Assets Question: Are new assets introduced in the paper well documented and is the documentation provided alongside the assets? Answer: [NA] Justification: the paper does not release new assets. 17 Guidelines: • The answer NA means that the paper does not release new asset...

  45. [54]

    Guidelines: • The answer NA means that the paper does not involve crowdsourcing nor research with human subjects

    Crowdsourcing and Research with Human Subjects Question: For crowdsourcing experiments and research with human subjects, does the paper include the full text of instructions given to participants and screenshots, if applicable, as well as details about compensation (if any)? A...

  46. [55]

    Guidelines: • The answer NA means that the paper does not involve crowdsourcing nor research with human subjects

    Institutional Review Board (IRB) Approvals or Equivalent for Research with Human Subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.