Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Pruning for Sparse Diffusion Models based on Gradient Flow

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The paper argues that iteratively pruning a diffusion model with a continuously relaxed mask, guided by a Hessian-gradient importance score, keeps generated images closer to the unpruned model than one-shot pruning.

desk verdict Combines two known pruning ideas into a new empirical study; the results are modest but plausible, with one load-bearing implementation detail left out. read the letter →

arxiv 2501.09464 v1 pith:QBRS2F2J submitted 2025-01-16 cs.LG

Ben Wan , Tianyi Zheng , Zhaoyu Chen , Yuxiao Wang , Jia Wang This is my paper · ORCID
classification cs.LG
keywords diffusionmodelsnetworkpruningprogressivesoftgradientflowHessian-gradientproductDDPMmodelcompression
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

Diffusion models generate high-quality images but run slowly, and existing one-shot pruning methods remove too much information at once. This paper argues that pruning should be gradual: a soft, continuous mask is tightened over many iterations, and each pruning decision is scored by how removing a weight changes the gradient norm of the diffusion loss. The authors report that on CIFAR-10, CelebA-HQ, LSUN-Church, and LSUN-Bedroom, this schedule beats magnitude, Taylor, and Diff-Pruning baselines at the same 50% sparsity and the same number of training steps, with better FID and SSIM. The practical claim is that a diffusion model can be compressed to half its size without the usual quality collapse.

What carries the argument

The central object is the continuous mask matrix $M$ with entries in $\{p_t,1\}$, where $p_t$ anneals from 1 to 0, paired with the gradient-flow importance score $I_t(\theta \odot M) = \theta \odot M \odot H_t g_t$. This score estimates the first-order change in the loss's gradient norm when a weight is removed, so weights whose removal would shrink the gradient flow are pruned first. The mask update is then described in Eq. (8) as gradient flow of the energy function in sparse space, which is the paper's mechanism for avoiding abrupt information loss while finding a mask that enables fast convergence.

What would settle it

Reproduce Algorithm 1 on CIFAR-10 with the same pretrained DDPM and 50% sparsity, but compute $H_t g_t$ with a single-sample empirical Fisher per timestep instead of the paper's unspecified estimator; if the FID gap over Diff-Pruning disappears, the criterion's advantage depends entirely on the unstated Hessian estimator.

Watch

Extended reading notes

Core claim

At its core, the paper claims that the mask used to prune a diffusion model should be optimized in a continuous space under a gradient-flow objective, rather than switched from 1 to 0 in one step. For the noise-prediction loss of a DDPM, it defines the importance of each parameter as $I_t(\theta \odot M) = \theta \odot M \odot H_t g_t$, where $H_t$ is the Hessian and $g_t$ the gradient of the loss at step $t$. Pruning the lowest-importance parameters while the mask value $p_t$ and sparsity $s_t$ are linearly annealed toward hard sparsity preserves the gradient flow, which the authors connect to faster convergence during iterative fine-tuning. The reported gains are concrete: on CIFAR-10 at 50% sparsity, FID improves from 5.29 (Diff-Pruning) to 5.18 and SSIM from 0.932 to 0.940, with similar improvements on the other three datasets. The paper's equation (8) formalizes the mask update as gradient flow in sparse space, but the load-bearing evidence is the empirical comparison.

Load-bearing premise

The importance ranking relies on the Hessian-gradient product $H_t g_t$ of the diffusion loss, yet the paper never states how this product is computed (exact Hessian, empirical Fisher, or another approximation) or how noise timesteps are aggregated, so an unstable estimate could invalidate the reported improvements.

Editorial extensions

If this is right

  • At 50% sparsity, the pruned CIFAR-10 model runs at 19.8M parameters and 3.4G MACs while reaching FID 5.18, within about one FID point of the 4.19 of the unpruned model, so the authors' schedule offers a practical compression path for DDPMs.
  • Because the same loss function drives both pruning and fine-tuning, the method composes with existing diffusion fine-tuning pipelines and needs no auxiliary losses such as FID-based criteria.
  • The progressive schedule should scale to deeper or latent diffusion models that currently rely on one-shot structural pruning, since the mask is only a continuous relaxation of the existing architecture.
  • The importance score directly targets convergence speed, so the method's benefit should be largest in the low-step fine-tuning regime, which is where diffusion deployment typically operates.

Reading between the lines

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

  • The paper does not verify Eq. (8) empirically; a reader could test whether the mask trajectory actually tracks the gradient of the energy, for example by measuring the cosine similarity between successive mask updates and the negative energy gradient.
  • The unstated Hessian estimation is the main transfer risk: if $H_t g_t$ is replaced by the gradient alone, the method should degrade toward Taylor-style pruning, which would isolate how much of the gain comes from the Hessian term.
  • The same progressive-soft-mask schedule could be applied to other generative models with a noise-prediction objective, such as latent diffusion, but the benefit would depend on whether their Hessian-gradient products remain informative at scale.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes a structured pruning method for pre-trained diffusion models that combines progressive soft pruning with a gradient-flow-based importance criterion. The mask is made continuous in [p_t,1], the target sparsity and mask value are scheduled linearly over the pruning phase, and weights are pruned according to an importance score θ⊙M⊙H_tg_t inspired by GraSP, followed by iterative pruning and fine-tuning. Experiments on CIFAR-10, CelebA-HQ, LSUN-Church, and LSUN-Bedroom report small improvements in FID and SSIM over one-shot magnitude, Taylor, and Diff-Pruning baselines at matched sparsity and training steps, and ablations attribute gains to soft pruning, progressive pruning, and the gradient-flow criterion.

Significance. If the empirical results hold, the paper offers a modest but consistent improvement in pruning quality for DDPMs, with a simple schedule that is easy to implement. The claimed theoretical contribution, however, is not substantiated: the derivation of the sparse-space gradient flow is asserted rather than proved, and the core importance criterion is left underspecified for the stochastic diffusion loss. The paper does not provide code, and several implementation details needed to reproduce the ranking are missing. On balance, the empirical comparison is a useful incremental data point, but the current manuscript does not yet establish the gradient-flow interpretation or a fully reproducible method.

major comments (4)
  1. [II-E, Eq. (8)] The equality ∆E(M) = ||1 − p_t(1 − I_{A_{s_t}}(I_t/||I_t||_2))||^2 is asserted without derivation. The left-hand side is the squared norm of the gradient of an energy function with respect to M, while the right-hand side is a norm of the mask update produced by the progressive soft-pruning rule. No mathematical link is given between ∇E(M) and the indicator-based update, and the equation is not implied by the preceding definitions. Since the title and abstract rest on the gradient-flow interpretation, this step is load-bearing; it should either be derived rigorously or the claim should be weakened to a heuristic analogy.
  2. [II-D, Eq. (7)] The quantity H_t g_t in Eq. (7) is never defined for the stochastic noise-prediction loss in Eq. (1). That loss is an expectation over x0, timestep t, and noise ε, and the paper does not state whether the Hessian-gradient product is computed by exact double backpropagation, an empirical Fisher approximation, a diagonal or block-diagonal Hessian, or another estimator, nor how timesteps and noise samples are aggregated. Because Algorithm 1 and all reported results depend on this importance ranking, the missing estimator specification prevents reproduction and makes the reported gains difficult to assess. In addition, the sign convention is unclear: Eq. (6) defines I_t(δ) for a perturbation δ, while Eq. (7) is written as θ⊙M⊙H_tg_t, and the text says lower I_t weights are removed while also saying removal should increase gradient norm.
  3. [Algorithm 1 and II-C] The mask update rule is underspecified. Line 8, 'Soft prune s_t of weights in eθ with p_t mask matrix based on the importance Eq.7,' does not state which entries become p_t and which become 1, how the continuous sparsity function G(M) = 1 − ||M − p_t||_0/(d_1d_2) is enforced during optimization, or whether the weights themselves are updated or reset between pruning rounds. A precise algorithmic specification of the mask update at each iteration is needed for reproducibility.
  4. [III-B, Table I] The paper states a target sparsity of s = 0.5, but the reported #Params ratios are about 55% of the pretrained model (e.g., 19.8M/35.7M on CIFAR-10 and 43.7M/78.1M on CelebA-HQ), corresponding to roughly 44–45% sparsity. This discrepancy should be clarified: does s refer to pruned channels, FLOPs, or another convention? Additionally, SparseDM [17], which is cited as the closest progressive pruning method for diffusion models, is not included in Table I or Table II; given that the paper's method is explicitly a progressive pruning method, the omission of this direct baseline leaves the comparative claim incomplete. The authors should either add SparseDM to the comparison or explain why it cannot be included.
minor comments (5)
  1. [III-A] There is a typo in 'Datesets' (should be 'Datasets').
  2. [References] Several references appear to have mismatched titles or authors (e.g., references [4], [5], [10], [11], and [12] do not correspond to the cited titles). Please verify all entries against the published versions.
  3. [Fig. 2] Figure 2 lacks axis labels, error bars, and a description of the pruning iterations plotted; adding these would make the convergence claim more interpretable.
  4. [III-A] The evaluation protocol is under-specified: the paper should state the number of FID evaluation samples, the reference statistics used, the random seeds, and the exact fine-tuning hyperparameters so that the comparisons are reproducible.
  5. [II-D, Eq. (5)] The term 'gradient flow' is used loosely; Eq. (5) is the directional derivative of the loss along its own gradient, not the gradient-flow differential equation dθ/dτ = −∇L(θ). Clarifying this terminology would avoid confusion with the standard notion of gradient flow.

Circularity Check

1 steps flagged · score 2.0 of 10

Empirical pruning comparisons are self-contained; only the 'gradient flow in sparse space' claim in Eq. 8 restates the mask update by construction.

  1. self definitional [Section II-E, Eq. (8)]
    "Based on the framework of our approach, we derive the gradient flow of the energy function adopting progressive soft pruning strategy: ∆E(M) = ∇E ⊤∇E = ||1 − pt (1 − IAst (It/||It||2))||2, (8) where I is indicator function and Ast is the value of top (1 − st)d-th largest normalized importance. The gradient flow in sparse space makes M can find the better mask matrix."

    The right-hand side of Eq. 8 is exactly the mask update implemented in Algorithm 1: the lowest-importance entries are set to pt and the rest are set to 1, so the vector whose squared norm is labeled ΔE(M) is just the algorithm's per-iteration mask change. No energy function E(M) is defined independently of this update rule, so the 'derived' gradient flow is constructed to equal the chosen mask update rather than predicting it. The gradient-flow interpretation in sparse space is therefore a restatement of the progressive soft-pruning procedure, not an independent derivation. This is a narrative circularity and does not by itself invalidate the external FID/SSIM comparisons in Tables I and II.

full rationale

The central empirical claim is checked against external baselines (magnitude pruning, Taylor pruning, Diff-Pruning) and against the pre-trained model, with fixed linear schedules for pt and st and no parameter fitted to the reported FID/SSIM values. The pruning criterion in Eq. 7 is adopted from the external GraSP work [19], not from the authors' own prior results, and the self-citations [24], [25] are not load-bearing for the pruning claim. The only genuine circularity is Eq. 8, where the 'gradient flow of the energy function' is defined to be the algorithm's own mask update; this is a self-definitional restatement rather than a predictive derivation. The unstated Hessian-gradient estimator for Eq. 7 is a reproducibility concern, not a circularity, because the ranking is computed from the diffusion loss rather than from the reported quality metrics. Overall, the empirical pruning comparison stands on its own, so the circularity score is low.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

No invented entities. The free parameters are hand-chosen hyperparameters with no sensitivity analysis. The axioms record the standard Taylor expansion, the unverified transfer of GraSP to diffusion, the missing Hessian specification, the ad hoc identification of the mask update with gradient flow, and the empirical assumption that progressive schedules help.

free parameters (4)
  • target sparsity s = 0.5
    Set to 50% for all experiments; chosen by hand as the compression target, no sensitivity analysis.
  • progressive soft pruning ratio N/K = 0.1
    Progressive phase set to 10% of total training iterations; no ablation on this ratio.
  • iterative pruning ratio M/K = 0.2
    Iterative pruning phase set to 20% of total training iterations; no ablation on this ratio.
  • mask value schedule p_t = 1 - t/N = linear from 1 to 0
    The soft mask value decreases linearly over the progressive phase; alternative schedules are not explored.
assumptions (5)
  • standard math The first-order Taylor expansion in Eq.6, with negligible O(||δ||^2), is valid for the pruning perturbation.
    Used to derive the GraSP importance in Eq.7; standard but unverified for large discrete mask changes in iterative pruning.
  • domain assumption The importance score I_t = θ⊙M⊙H_t g_t computed on the current pruned network predicts which weights to remove in later iterations.
    Core transfer assumption imported from GraSP [19], invoked in Section II.D without re-derivation for diffusion models.
  • domain assumption The Hessian-gradient product H_t g_t for the diffusion loss can be computed or approximated, and the approximation does not change the ranking.
    No implementation detail for H is given; results depend on this unstated approximation.
  • ad hoc to paper The energy function E(M) is differentiable in the soft mask M over [p,1], and its gradient flow equals the mask update rule in Eq.8.
    Eq.8 asserts this equality with no derivation; this is an ad hoc construction supporting the 'gradient flow in sparse space' narrative.
  • domain assumption Linearly ramping s_t and p_t (progressive soft pruning) improves final FID and SSIM relative to one-shot pruning.
    Supported empirically in Table II on two datasets, but treated as a general principle without theory.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Pruning for Sparse Diffusion Models based on Gradient Flow." pith.science (2026). https://pith.science/paper/QBRS2F2J

@misc{pith2026250109464,
  author       = {Pith},
  title        = {Pith review of: Pruning for Sparse Diffusion Models based on Gradient Flow},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QBRS2F2J}},
  note         = {Machine review of arXiv:2501.09464}
}
read the original abstract

Diffusion Models (DMs) have impressive capabilities among generation models, but are limited to slower inference speeds and higher computational costs. Previous works utilize one-shot structure pruning to derive lightweight DMs from pre-trained ones, but this approach often leads to a significant drop in generation quality and may result in the removal of crucial weights. Thus we propose a iterative pruning method based on gradient flow, including the gradient flow pruning process and the gradient flow pruning criterion. We employ a progressive soft pruning strategy to maintain the continuity of the mask matrix and guide it along the gradient flow of the energy function based on the pruning criterion in sparse space, thereby avoiding the sudden information loss typically caused by one-shot pruning. Gradient-flow based criterion prune parameters whose removal increases the gradient norm of loss function and can enable fast convergence for a pruned model in iterative pruning stage. Our extensive experiments on widely used datasets demonstrate that our method achieves superior performance in efficiency and consistency with pre-trained models.

Figures

Figures reproduced from arXiv: 2501.09464 by the authors.

Figure 1
Figure 1. Pruning for diffusion models based on gradient flow, including the gradient flow pruning process and the gradient flow pruning criterion. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FID change in iterative pruning stage using progressive soft pruning. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. DiTReducio: A Training-Free Acceleration for DiT-Based TTS via Progressive Calibration

    cs.SD 2025-09 conditional novelty 6.0 of 10

    DiTReducio is a training-free, pattern-guided layer and branch skipping method that accelerates DiT-based TTS, reporting significant FLOP and RTF reductions with modest quality loss at tuned thresholds.

Reference graph

Works this paper leans on

35 extracted references · 22 canonical work pages · cited by 1 Pith paper

  1. [17]

    ”SparseDM: Toward Sparse Efficient Diffusion Models.” arXiv preprint arXiv:2404.10445 (2024)

    Wang, Kafeng, et al. ”SparseDM: Toward Sparse Efficient Diffusion Models.” arXiv preprint arXiv:2404.10445 (2024)

    arXiv 2024

  2. [1]

    ”Denoising Diffusion Prob- abilistic Models.” Advances in Neural Information Processing Systems (NeurIPS), 2020

    Ho, Jonathan, Ajay Jain, and Pieter Abbeel. ”Denoising Diffusion Prob- abilistic Models.” Advances in Neural Information Processing Systems (NeurIPS), 2020

    work page 2020

  3. [2]

    ”Improved Denoising Diffusion Probabilistic Models.” arXiv preprint arXiv:2102.09672, 2021

    Nichol, Alexander Q., and Prafulla Dhariwal. ”Improved Denoising Diffusion Probabilistic Models.” arXiv preprint arXiv:2102.09672, 2021

    arXiv 2021

  4. [3]

    Kingma, Abhishek Ku- mar, Stefano Ermon, and Ben Poole

    Song, Yang, Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Ku- mar, Stefano Ermon, and Ben Poole. ”Score-Based Generative Modeling through Stochastic Differential Equations.” International Conference on Learning Representations (ICLR), 2021

    work page 2021

  5. [4]

    Higher angular momentum pairings in interorbital shadowed-triplet superconductors: Application to Sr$_{2}$RuO$_{4}$

    Carlini, Nicholas, Milad Nasr, Tim Salimans, and Shumeng Song. ”Guided Diffusion Models for Adversarial Purification and Image Gen- eration.” arXiv preprint arXiv:2107.00047, 2021

    work page Pith review arXiv 2021

  6. [5]

    The structure of the unit group of the group algebra $F(C_3 \times D_{10})$

    Nichol, Alexander Q., and Prafulla Dhariwal. ”Learning in Latent Space with Diffusion Models.” arXiv preprint arXiv:2106.02265, 2021

    work page Pith review arXiv 2021

  7. [6]

    Weiss, Mohammad Norouzi, and William Chan

    Chen, Naihan, Zehua Zhang, Heiga Zen, Ron J. Weiss, Mohammad Norouzi, and William Chan. ”WaveGrad: Estimating Gradients for Waveform Generation.” International Conference on Learning Repre- sentations (ICLR), 2021

    work page 2021

  8. [7]

    ”DiffWave: A Versatile Diffusion Model for Audio Synthesis.” International Conference on Learning Representations (ICLR), 2021

    Kong, Zhifeng, Wei Ping, Jiaji Huang, Kexin Zhao, and Bryan Catan- zaro. ”DiffWave: A Versatile Diffusion Model for Audio Synthesis.” International Conference on Learning Representations (ICLR), 2021

    work page 2021

Show all 35 references
  1. [8]

    ”Diffusion-LM Improves Controllable Text Generation.” arXiv preprint arXiv:2205.14217 (2022)

    Li, Yifan, et al. ”Diffusion-LM Improves Controllable Text Generation.” arXiv preprint arXiv:2205.14217 (2022)

    2022 arXiv

  2. [9]

    ”DiffuSeq: Sequence to Sequence Text Generation with Diffusion Models.” arXiv preprint arXiv:2302.04023 (2023)

    Gong, Chengxi, et al. ”DiffuSeq: Sequence to Sequence Text Generation with Diffusion Models.” arXiv preprint arXiv:2302.04023 (2023)

    2023 arXiv

  3. [10]

    ”Elucidating the Design Space of Diffusion-Based Generative Models.” arXiv preprint arXiv:2105.05233 (2021)

    Nichol, Alexander Quinn, and Prafulla Dhariwal. ”Elucidating the Design Space of Diffusion-Based Generative Models.” arXiv preprint arXiv:2105.05233 (2021)

    2021 arXiv

  4. [11]

    ”Fast Sampling of Diffusion Models with Exponential Integrator.” Advances in Neural Information Processing Systems (NeurIPS), 2021

    Song, Jiaming, Lantao Yu, and Stefano Ermon. ”Fast Sampling of Diffusion Models with Exponential Integrator.” Advances in Neural Information Processing Systems (NeurIPS), 2021

    2021

  5. [12]

    ”Learning Fast Samplers for Diffusion Models by Differentiating Through Sample Quality.” arXiv preprint arXiv:2204.07166 (2022)

    De Cao, Nicola, Wilker Aziz, and Ivan Titov. ”Learning Fast Samplers for Diffusion Models by Differentiating Through Sample Quality.” arXiv preprint arXiv:2204.07166 (2022)

    2022 arXiv

  6. [13]

    ”Kdgan: Knowledge distillation with generative ad- versarial networks.” Advances in neural information processing systems 31 (2018)

    Wang, Xiaojie, et al. ”Kdgan: Knowledge distillation with generative ad- versarial networks.” Advances in neural information processing systems 31 (2018)

    2018

  7. [14]

    ”Diffusion Mod- els for Text Generation without Exposure Bias.” arXiv preprint arXiv:2301.11309 (2023)

    Laidlaw, William, and Anastasios Angelopoulos. ”Diffusion Mod- els for Text Generation without Exposure Bias.” arXiv preprint arXiv:2301.11309 (2023)

    2023 arXiv

  8. [15]

    ”Denoising diffusion implicit models.” arXiv preprint arXiv:2010.02502 (2020)

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

    2020 arXiv

  9. [16]

    ”Structural Pruning for Diffusion Models.” Advances in Neural Information Processing Systems (NeurIPS), 2023

    Wang, Gongfan Fang Xinyin Ma Xinchao. ”Structural Pruning for Diffusion Models.” Advances in Neural Information Processing Systems (NeurIPS), 2023

    2023

  10. [18]

    Optimal transport: old and new

    Villani, C ´edric. Optimal transport: old and new. V ol. 338. Berlin: springer, 2009

    2009

  11. [19]

    ”Picking winning tickets before training by preserving gradient flow.” arXiv preprint arXiv:2002.07376 (2020)

    Wang, Chaoqi, Guodong Zhang, and Roger Grosse. ”Picking winning tickets before training by preserving gradient flow.” arXiv preprint arXiv:2002.07376 (2020)

    2020 arXiv

  12. [20]

    ”Soft masking for cost-constrained channel prun- ing.” European Conference on Computer Vision

    Humble, Ryan, et al. ”Soft masking for cost-constrained channel prun- ing.” European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022

    2022

  13. [21]

    ”Training recipe for n: M structured sparsity with decaying pruning mask.” arXiv preprint arXiv:2209.07617 (2022)

    Kao, Sheng-Chun, et al. ”Training recipe for n: M structured sparsity with decaying pruning mask.” arXiv preprint arXiv:2209.07617 (2022)

    2022 arXiv

  14. [22]

    ”Learning n: m fine-grained structured sparse neural networks from scratch.” arXiv preprint arXiv:2102.04010 (2021)

    Zhou, Aojun, et al. ”Learning n: m fine-grained structured sparse neural networks from scratch.” arXiv preprint arXiv:2102.04010 (2021)

    2021 arXiv

  15. [23]

    Lubana, Ekdeep Singh, and Robert P. Dick. ”A gradient flow frame- work for analyzing network pruning.” arXiv preprint arXiv:2009.11839 (2020)

    2020 arXiv

  16. [24]

    T., Wan, B., Zhang, H., Chen, J.,

    Zheng, T., Geng, C., Jiang, P. T., Wan, B., Zhang, H., Chen, J., ... & Li, B. (2024, October). Non-uniform timestep sampling: Towards faster diffusion model training. In Proceedings of the 32nd ACM International Conference on Multimedia (pp. 7036-7045)

    2024

  17. [25]

    T., Wan, B., Zhang, H., Chen, J., Wang, J., & Li, B

    Zheng, T., Jiang, P. T., Wan, B., Zhang, H., Chen, J., Wang, J., & Li, B. (2025). Beta-tuned timestep diffusion model. In European Conference on Computer Vision (pp. 114-130). Springer, Cham

    2025

  18. [26]

    Chen, Z., Li, B., Wu, S., Jiang, K., Ding, S., & Zhang, W. (2024). Content-based unrestricted adversarial attack. Advances in Neural In- formation Processing Systems, 36

    2024

  19. [27]

    ”Only train once: A one-shot neural network training and pruning framework.” Advances in Neural Information Processing Systems 34 (2021): 19637-19651

    Chen, Tianyi, et al. ”Only train once: A one-shot neural network training and pruning framework.” Advances in Neural Information Processing Systems 34 (2021): 19637-19651

    2021

  20. [28]

    ”The lottery ticket hy- pothesis: Finding sparse, trainable neural networks.” arXiv preprint arXiv:1803.03635 (2018)

    Frankle, Jonathan, and Michael Carbin. ”The lottery ticket hy- pothesis: Finding sparse, trainable neural networks.” arXiv preprint arXiv:1803.03635 (2018)

    2018 arXiv

  21. [29]

    ”Learning multiple layers of features from tiny images.” (2009): 7

    Krizhevsky, Alex, and Geoffrey Hinton. ”Learning multiple layers of features from tiny images.” (2009): 7

    2009

  22. [30]

    ”Progressive Growing of GANs for Improved Quality, Stability, and Variation.” arXiv preprint arXiv:1710.10196 (2017)

    Karras, Tero. ”Progressive Growing of GANs for Improved Quality, Stability, and Variation.” arXiv preprint arXiv:1710.10196 (2017)

    2017 arXiv

  23. [31]

    ”Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop.” arXiv preprint arXiv:1506.03365 (2015)

    Yu, Fisher, et al. ”Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop.” arXiv preprint arXiv:1506.03365 (2015)

    2015 arXiv

  24. [32]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium

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

    2017

  25. [33]

    Bovik, Hamid R

    Zhou Wang, Alan C. Bovik, Hamid R. Sheikh, and Eero P. Simoncelli, ”Image quality assessment: from error visibility to structural similairty,” IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600-612, Apr. 2004

    2004

  26. [34]

    Channel pruning for accelerat- ing very deep neural networks

    Yihui He, Xiangyu Zhang, and Jian Sun. Channel pruning for accelerat- ing very deep neural networks. In Proceedings of the IEEE international conference on computer vision, pages 1389–1397, 2017

    2017

  27. [35]

    Pruning convolutional neural networks for resource efficient inference

    Pavlo Molchanov, Stephen Tyree, Tero Karras, Timo Aila, and Jan Kautz. Pruning convolutional neural networks for resource efficient inference. arXiv preprint arXiv:1611.06440, 2016

    2016 arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.