Pith. sign in

REVIEW 4 major objections 6 minor 67 references

Nested Annealed Training Scheme for Generative Adversarial Networks

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A nested annealed training scheme gives any GAN the generator dynamics of annealed CFG and reports lower FID and higher Inception Score across model families.

desk verdict A concrete GAN training recipe with promising-looking FID numbers, but the central theory (Prop 4) is unsupported and as written appears to conflate two different gradient objects. read the letter →

arxiv 2501.11318 v1 pith:OPLYTQPK submitted 2025-01-20 cs.CV cs.LG

Chang Wan , Ming-Hsuan Yang , Minglu Li , Yunliang Jiang , Zhonglong Zheng This is my paper · ORCID
classification cs.CVcs.LG
keywords generativeadversarialnetworksscore-basedmodelannealedweightcompositefunctionalgradientnestedtrainingschemeimagegenerationFréchetInceptionDistanceScore
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

This paper tries to give GAN training a score-function foundation. It argues that in the composite-functional-gradient GAN (CFG), the optimal discriminator's gradient equals the difference between the score functions of real and synthesized samples, and that the generator is best seen as minimizing that score difference. From score-based model annealing, it derives a decreasing weight on the discriminator gradient, calling the result annealed CFG, and then packages that weight into a nested training loop (NATS) that can be inserted into existing GANs without changing their loss or architecture. If the paper's central equivalence holds, NATS would let any GAN inherit annealed CFG's generator dynamics and improve sample quality and diversity. The experiments report lower FID and higher Inception Score from the original GAN through BigGAN and DDGAN.

What carries the argument

The load-bearing object is the annealed weight vector $w$, a decreasing sequence of scalars that multiplies discriminator gradients in annealed CFG and sits inside the generator loss in NATS. The nested training loop is the second piece: by training the discriminator once and the generator $j$ times at step $j$, it creates a sub-loop in which accumulated discriminator gradients can guide the generator, which the common one-step alternating scheme lacks. Proposition 4, claiming equality of the NATS and annealed CFG gradient-vector fields, is what lets the annealed weight transfer across architectures, losses, and regularizations.

What would settle it

Compute the generator gradient of the NATS objective and compare it numerically to the annealed CFG gradient during an actual training run: the NATS gradient is $-w[j] D'(G(z))/(1-w[j]D(G(z))) \cdot \partial G/\partial\theta_g$, whereas the annealed CFG gradient is $w_m(x)\nabla_x D_m(x)\cdot\partial G/\partial\theta_g$. If the two vector fields diverge in direction or magnitude, Proposition 4 fails; if they match, the claim is confirmed.

Watch

Extended reading notes

Core claim

The central claim is that the CFG discriminator's gradient $\nabla_x D(x)$ is $\nabla_x \log p^*(x) - \nabla_x \log p_g(x)$, the difference of the score functions of real and synthesized data, and that the CFG generator's objective is minimized when this difference vanishes. On this basis the paper introduces an annealed weight $w_m(x)$ in front of each discriminator gradient, inspired by annealed Langevin dynamics, and names the result annealed CFG. To make the idea usable with modern GANs, it proposes NATS: a nested scheme where the discriminator is updated once and the generator several times, with the generator objective $\log(1 - w[j]D(G(z)))$ carrying the weight. Proposition 4 states that the NATS generator has the same gradient-vector field as annealed CFG, so the annealing benefit transfers to any GAN that adopts the scheme.

Load-bearing premise

The load-bearing premise is that the NATS generator update, which places the annealed weight $w[j]$ inside $\log(1 - w[j]D(G(z)))$, yields exactly the same gradient-vector field as annealed CFG, which scales the input gradient $\nabla_x D_m(x)$ by $w_m(x)$; the proof of this equality is deferred to an appendix that is absent from the preprint.

Editorial extensions

If this is right

  • If Proposition 4 is correct, any GAN trained with NATS reproduces annealed CFG's generator dynamics without redesigning the model, so the annealed-training benefit becomes architecture-agnostic.
  • The scheme should keep working when the loss changes, since the modification is confined to the generator update and the annealed weight.
  • On the paper's empirical results, NATS lowers CIFAR-10 FID for all tested GAN families and improves the state-of-the-art baselines: BigGAN from 8.52 to 6.36 and DDGAN from 3.75 to 2.56.
  • The reverse-annealing experiment in the paper supports the direction of the weight schedule: reversing $w$ worsens FID, consistent with the annealing rationale.
  • The score-function identity gives GAN training a diagnostic handle: the discriminator gradient is interpretable as the mismatch between the real-data score and the generator's score.

Reading between the lines

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

  • Editorial inference: if the score-difference identity is the mechanism behind the gains, then NATS training should visibly shrink the distance between the generator's score and the data score; measuring both scores during training would test that mechanism directly.
  • Editorial inference: the nested loop is itself a training-schedule change, so the paper's NTS-versus-NATS comparison can be read as separating the benefit of annealing from the benefit of extra generator steps per discriminator update.
  • Editorial inference: putting $w[j]$ inside the logarithm makes the weight interact with the discriminator output before the gradient is taken, while annealed CFG scales the input-gradient directly; a direct comparison of these two vector fields would show whether the equality claimed in Proposition 4 is exact or only approximate.
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 / 6 minor

Summary. The paper claims to establish a theoretical connection between the composite-functional-gradient GAN (CFG) and score-based models, to derive an annealed weighting of the CFG discriminator gradient, and to introduce a nested annealed training scheme (NATS) that transfers this annealing to arbitrary GAN losses. The central theoretical claim is Proposition 4, which states that the generator update in NATS has the same gradient-vector field as the annealed CFG method. The paper reports FID and IS improvements on CIFAR-10, LSUN, CelebA, and ImageNet64 across several GAN families, including BigGAN and DDGAN.

Significance. If Proposition 4 were true, the paper would offer a principled way to import annealed weighting from score-based models into any GAN, with broad practical value. The paper should be credited for Proposition 1, which is a correct and clean restatement of the optimal CFG discriminator gradient as a difference of score functions, and for a broad experimental comparison across original GAN, LSGAN, WGAN, HingeGAN, BigGAN, and DDGAN, including an ablation over N_d. However, the contribution rests on Proposition 4, and that proposition is not established: its proof is deferred to a missing appendix, and the algorithm-level gradient objects in Algorithms 2 and 3 are different. The empirical gains, while suggestive, are conditional on dataset-specific hyperparameter choices and are reported without statistical uncertainty.

major comments (4)
  1. [III-D, Proposition 4; Algorithm 3 line 10 vs Algorithm 2 lines 5-6] The load-bearing claim is not supported and, as stated, is inconsistent with the printed algorithms. In annealed CFG, the generator is updated in data space by G_{m+1}(z) = G_m(z) + η_m w_m(x) ∇_x D_m(x), whereas Algorithm 3 updates generator parameters θ_g by ascending ∇_θ (1/m) Σ_i log(1 - w[j] D(G(z_i))). Differentiating the latter yields terms of the form - w[j] D'(G(z)) G_θ(z) / (1 - w[j] D(G(z))), which contain a denominator and a Jacobian structure that have no counterpart in Algorithm 2's scaled input-gradient accumulation. The text refers to Appendix Section B for the proof of Proposition 4, but no appendix is present in the preprint. The claimed equality of gradient-vector fields therefore requires either a corrected algorithm or a nontrivial identity that the manuscript does not provide.
  2. [III-C, Propositions 2 and 3] These propositions are stated without proof, and Proposition 3's notation is not well-defined: Eq. (8) uses δ^(i)_m, r^(i)_m, and D^(i)_m without specifying what i indexes or how N relates to the M accumulation steps of Eq. (1), and the claim that w_m(x) = δ_m(x) r_m(x) is asserted rather than derived. Since Eq. (8) is the basis for the annealed weight and for Proposition 4, this gap is important.
  3. [III-C, Initial Annealed Weight; Tables I and V] The annealed weight schedule is not derived from first principles: Technique 1 sets w_1 from the median pairwise distance D in Table I, and the remaining r_m are chosen as a geometric progression with a ratio set empirically; Table V then fixes different w(x) vectors for low- and high-resolution data. The reported improvements are therefore conditional on hyperparameters that are fitted to the evaluation datasets. The paper should report the selection procedure and its sensitivity more completely, for example through cross-validation or a fixed rule, before claiming that NATS significantly improves SOTA GANs.
  4. [IV-B, Tables IX-XII] The empirical claims are reported without standard deviations, number of seeds, or statistical tests, despite the use of the word 'significant'. Table X is also confusing: the CFG row lists Reverse NATS = 200.01, NATS = 16.34, NTS = 19.41, and CTS = 19.41, but NATS, NTS, and CTS are defined for GAN training schemes, and it is unclear how they apply to CFG. This weakens the support for the headline empirical claim.
minor comments (6)
  1. [Author affiliation line] The affiliation line contains a typo: 'Minglu Li, Fellow, IEEE,,' has a double comma.
  2. [Section III-A and general references] The text repeatedly refers to Appendix Sections B, C, D, and E, but none of these appendices appears in the preprint, making it impossible to verify the cited proofs and additional results.
  3. [Algorithm 2, line 6] The update writes η_m g(G_m(z)) where the subscript on g is missing; it should be g_m to match Eq. (1).
  4. [Proposition 2] The notation 'z ∼ N(x | 0, I)' appears to mean z ∼ N(0, I), and the double summation over i and m is notationally unclear.
  5. [Notation for w] The text uses w(x), w_m(x), w[m], and w[j] for closely related quantities without a consistent convention; this makes Algorithm 3, line 10, ambiguous because w[j] is a scalar while w(x) is described as a vector.
  6. [Tables VII and VIII] The metric is labeled 'Fr´echet Distance'; it should be 'Fr´echet Inception Distance' to match the text and the cited implementation.

Circularity Check

1 steps flagged · score 1.0 of 10

No significant construction-level circularity: the load-bearing Proposition 4 is unproved (proof in absent Appendix B) rather than definitionally forced; the only circular-adjacent item is Proposition 3, a definition presented as a theorem.

  1. self definitional [Section III-C, Proposition 3 (Eq. 8)]
    "Proposition 3. The analytic form of gm(x) in our annealed CFG model is expressed as follows: gm(x) = Σ_i δ_m^{(i)}(x) r_m^{(i)}(x) ∇_x D_m^{(i)}(x) = Σ_i w_m^{(i)}(x)∇_xD_m^{(i)}(x), where δ_m^{(i)}(x) refers to the hyper-parameters from CFG, and r_m^{(i)}(x) is a coefficient from the score-based model and w_m^{(i)}(x) = δ_m^{(i)}(x)r_m^{(i)}(x)."

    The asserted 'analytic form' is tautological: w_m is introduced as δ_m r_m, so the equation holds by construction rather than by derivation. The weight values are not obtained from this proposition; they are later fixed from the median-pairwise-distance Table I and an empirical geometric ratio γ≈0.5, after which FID/IS are measured. Thus the proposition is a definition dressed as a theorem. It is not used to force a prediction, so the circularity is minor and non-load-bearing.

full rationale

Proposition 1 is a direct corollary of D(x)=log(p*/pg) in Eq. (7), not circular. Proposition 2 restates the CFG generator's own update direction in terms of score differences; it feeds no target result back into itself. The annealed schedule is imported from the external NCSN geometric progression [14] plus Table I distances; the reported FID/IS are measured outcomes, not quantities derived from the schedule, so this is an empirical hyperparameter choice rather than a fitted-value-renamed-as-prediction. The load-bearing Proposition 4 is the real problem: it is asserted with the proof deferred to 'Appendix Section B,' which is absent from the preprint, and the printed Algorithm 3 generator update ∇θ log(1 - w[j]D(G(z))) is not visibly the same gradient-vector field as Algorithm 2's w_m(x)∇_xD_m(x). That is an unproved—and as written likely false—equivalence, but it is a validity/rigor defect, not a circularity, because nothing in Algorithms 2 or 3 forces the two fields to coincide by definition or by fitting. No load-bearing author self-citations are present. Overall, the only circular-adjacent element is Proposition 3, a definition presented as a proposition; it does not drive the empirical conclusions, so the paper has no significant construction-level circularity.

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

The central mechanism depends on four fitted or hand-set hyperparameters and on several assumptions about Gaussianity and transferability of score-based annealing to GAN gradients. Proposition 4, the load-bearing theoretical bridge, is an asserted axiom with its proof in a missing appendix.

free parameters (4)
  • w(x) annealed weight vector = [1, 0.01] for resolutions below 256; [20, 1] for 256x256
    Chosen per dataset/resolution from median pairwise distances D1/D2 (Table I, Technique 1) and empirical tuning; directly controls the generator update.
  • r_m(x) score-based coefficient = geometric progression with ratio approximately 0.5
    Borrowed from score-based model annealing, set by hand so w1 > w2 > ... > wM; no derivation given.
  • Nd nested loop length = 4 to 20 depending on GAN model
    Tuned per model for best FID/IS (Table VI, Table XII); the paper itself notes values must be adjusted per loss.
  • M CFG accumulation steps = 15
    Set for all CFG experiments without a search or theoretical basis.
assumptions (5)
  • standard math The optimal discriminator for the logistic loss is D*(x)=log(p*(x)/pg(x)).
    Used to derive Proposition 1; correct for the stated loss but standard in f-GAN theory.
  • domain assumption The synthetic distribution at generator step 1 is N(0,I) and the real data is Gaussian with covariance Sigma*.
    Equation (10) derives the initial annealed weight by assuming Gaussian forms for p_g1 and p*; real images are not Gaussian.
  • ad hoc to paper Score-based model annealing coefficients r_m can be transplanted onto CFG discriminator gradients.
    Section III-C asserts this without derivation; the resulting w_m is used as the central mechanism.
  • ad hoc to paper The generator update in Algorithm 3 has the same gradient-vector field as the annealed CFG update.
    Proposition 4; proof deferred to missing Appendix B and not evident from the algorithm.
  • domain assumption Regularity conditions guaranteeing existence of densities and gradients hold for image distributions.
    Needed for score functions and functional gradients; not verified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Nested Annealed Training Scheme for Generative Adversarial Networks." pith.science (2026). https://pith.science/paper/OPLYTQPK

@misc{pith2026250111318,
  author       = {Pith},
  title        = {Pith review of: Nested Annealed Training Scheme for Generative Adversarial Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OPLYTQPK}},
  note         = {Machine review of arXiv:2501.11318}
}
read the original abstract

Recently, researchers have proposed many deep generative models, including generative adversarial networks(GANs) and denoising diffusion models. Although significant breakthroughs have been made and empirical success has been achieved with the GAN, its mathematical underpinnings remain relatively unknown. This paper focuses on a rigorous mathematical theoretical framework: the composite-functional-gradient GAN (CFG)[1]. Specifically, we reveal the theoretical connection between the CFG model and score-based models. We find that the training objective of the CFG discriminator is equivalent to finding an optimal D(x). The optimal gradient of D(x) differentiates the integral of the differences between the score functions of real and synthesized samples. Conversely, training the CFG generator involves finding an optimal G(x) that minimizes this difference. In this paper, we aim to derive an annealed weight preceding the weight of the CFG discriminator. This new explicit theoretical explanation model is called the annealed CFG method. To overcome the limitation of the annealed CFG method, as the method is not readily applicable to the SOTA GAN model, we propose a nested annealed training scheme (NATS). This scheme keeps the annealed weight from the CFG method and can be seamlessly adapted to various GAN models, no matter their structural, loss, or regularization differences. We conduct thorough experimental evaluations on various benchmark datasets for image generation. The results show that our annealed CFG and NATS methods significantly improve the quality and diversity of the synthesized samples. This improvement is clear when comparing the CFG method and the SOTA GAN models.

Figures

Figures reproduced from arXiv: 2501.11318 by the authors.

Figure 2
Figure 2. The figure illustrates the concept derived from a score￾based model [14], where the score represents the gradient of the logarithmic density function of real samples. This indicates the direction through which noise samples evolve towards real samples. represents a descending order such that α1 > α2 > · · · > αL. For this reason, the method is called the annealed Langevin dynamics sampling. III. METHOD This section … view at source ↗
Figure 3
Figure 3. (a) A failure synthesis sample of CFG and gradient of discriminator in each M Step in 256×256 resolution LSUN Church. (b) Synthesized sample of Annealed CFG and gradient of discriminator in each M Step in 256×256 resolution LSUN Church The visual effect in the M = 1 column is more ambiguous than in the M = 15 column. In the CFG method, each M step shares the same gradient weight. When these gradients accumulate toge… view at source ↗
Figure 4
Figure 4. Best FID scores for various types of GAN archi￾tectures: original GAN, LSGAN, WGAN, and HingeGAN – under different settings for Nd. The values of Nd considered are 5, 10, 15, and 20. The red line represents the performance of NATS. The other colored lines represent the performance of NTS. B. Experimental Results In this section, we present a comparison of our Annealed CFG and NTS/NATS experimental among the CFG meth… view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Performance of BigGAN trained on the CIFAR10 dataset using different training schemes. Column (a): This column displays synthetic samples generated by BigGAN trained with the CTS. Column (b): Here, we present the synthetic samples from BigGAN when trained using the NTS…
Figure 6
Figure 6. Figure 6: We present the outcomes of a DDGAN trained on the CIFAR10 dataset using various training schemes. Column (a): This column displays the synthetic samples produced by the DDGAN when trained with the CTS. Column (b): Here, the synthetic samples from the DDGAN trained usin…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

67 extracted references · 44 canonical work pages

  1. [1]

    Composite functional gradient learning of generative adversarial models,

    R. Johnson and T. Zhang, “Composite functional gradient learning of generative adversarial models,” in International Conference on Machine Learning. PMLR, 2018, pp. 2371–2379

    work page 2018

  2. [2]

    Imagen video: High definition video generation with diffusion models,

    J. Ho, W. Chan, C. Saharia, J. Whang, R. Gao, A. Gritsenko, D. P. Kingma, B. Poole, M. Norouzi, D. J. Fleet et al. , “Imagen video: High definition video generation with diffusion models,” arXiv preprint arXiv:2210.02303, 2022

    arXiv

  3. [3]

    Zero-shot text-to-image generation,

    A. Ramesh, M. Pavlov, G. Goh, S. Gray, C. V oss, A. Radford, M. Chen, and I. Sutskever, “Zero-shot text-to-image generation,” in International Conference on Machine Learning . PMLR, 2021, pp. 8821–8831

    2021

  4. [4]

    Hierarchical text-conditional image generation with clip latents,

    A. Ramesh, P. Dhariwal, A. Nichol, C. Chu, and M. Chen, “Hierarchical text-conditional image generation with clip latents,” arXiv preprint arXiv:2204.06125, 2022

    arXiv 2022

  5. [5]

    Drag your gan: Interactive point-based manipulation on the generative image manifold,

    X. Pan, A. Tewari, T. Leimk ¨uhler, L. Liu, A. Meka, and C. Theobalt, “Drag your gan: Interactive point-based manipulation on the generative image manifold,” arXiv preprint arXiv:2305.10973 , 2023

    arXiv 2023

  6. [6]

    Ldm3d: Latent diffusion model for 3d,

    G. B. M. Stan, D. Wofk, S. Fox, A. Redden, W. Saxton, J. Yu, E. Aflalo, S.-Y . Tseng, F. Nonato, M. Muller et al. , “Ldm3d: Latent diffusion model for 3d,” arXiv preprint arXiv:2305.10853 , 2023

    arXiv 2023

  7. [7]

    Locally attentional sdf diffusion for controllable 3d shape generation,

    X.-Y . Zheng, H. Pan, P.-S. Wang, X. Tong, Y . Liu, and H.-Y . Shum, “Locally attentional sdf diffusion for controllable 3d shape generation,” arXiv preprint arXiv:2305.04461 , 2023

    arXiv 2023

  8. [8]

    Mm-diffusion: Learning multi-modal diffusion models for joint audio and video generation,

    L. Ruan, Y . Ma, H. Yang, H. He, B. Liu, J. Fu, N. J. Yuan, Q. Jin, and B. Guo, “Mm-diffusion: Learning multi-modal diffusion models for joint audio and video generation,” in IEEE/CVF Computer Vision and Pattern Recognition Conference, 2023, pp. 10 219–10 228

    work page 2023

Show all 67 references
  1. [9]

    Generative adversarial nets,

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial nets,” Conference on Neural Information Processing Systems , vol. 27, 2014

    2014

  2. [10]

    Statistics enhancement generative adversarial networks for diverse con- ditional image synthesis,

    Z. Zuo, A. Li, Z. Wang, L. Zhao, J. Dong, X. Wang, and M. Wang, “Statistics enhancement generative adversarial networks for diverse con- ditional image synthesis,” IEEE Transactions on Circuits and Systems for Video Technology, pp. 1–1, 2023

    2023

  3. [11]

    Mitigating label noise in gans via enhanced spectral normalization,

    Y . Chen, C. Jin, G. Li, T. H. Li, and W. Gao, “Mitigating label noise in gans via enhanced spectral normalization,” IEEE Transactions on Circuits and Systems for Video Technology , vol. 33, no. 8, pp. 3924– 3934, 2023

    2023

  4. [12]

    Hrinversion: High- resolution gan inversion for cross-domain image synthesis,

    P. Zhou, L. Xie, B. Ni, L. Liu, and Q. Tian, “Hrinversion: High- resolution gan inversion for cross-domain image synthesis,” IEEE Trans- actions on Circuits and Systems for Video Technology , vol. 33, no. 5, pp. 2147–2161, 2023

    2023

  5. [13]

    A framework of composite functional gradient methods for generative adversarial models,

    R. Johnson and T. Zhang, “A framework of composite functional gradient methods for generative adversarial models,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 43, no. 1, pp. 17–32, 2019

    2019

  6. [14]

    Generative modeling by estimating gradients of the data distribution,

    Y . Song and S. Ermon, “Generative modeling by estimating gradients of the data distribution,” Conference on Neural Information Processing Systems, vol. 32, 2019

    2019

  7. [15]

    Annealed importance sampling,

    R. M. Neal, “Annealed importance sampling,” Statistics and Computing, vol. 11, pp. 125–139, 2001

    2001

  8. [16]

    Optimization by simulated annealing,

    S. Kirkpatrick, C. D. Gelatt Jr, and M. P. Vecchi, “Optimization by simulated annealing,” Science, vol. 220, no. 4598, pp. 671–680, 1983

    1983

  9. [17]

    Learning multiple layers of features from tiny images,

    A. Krizhevsky, G. Hinton et al. , “Learning multiple layers of features from tiny images,” Master’s thesis, Department of Computer Science, University of Toronto, 2009

    2009

  10. [18]

    Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop,

    F. Yu, A. Seff, Y . Zhang, S. Song, T. Funkhouser, and J. Xiao, “Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop,” arXiv preprint arXiv:1506.03365 , 2015

    2015 arXiv

  11. [19]

    Deep learning face attributes in the wild,

    Z. Liu, P. Luo, X. Wang, and X. Tang, “Deep learning face attributes in the wild,” in InternationalConference on Computer Vision, December 2015

    2015

  12. [20]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in IEEE/CVF Computer Vision and Pattern Recognition Conference . Ieee, 2009, pp. 248–255

    2009

  13. [21]

    Improved training of wasserstein gans,

    I. Gulrajani, F. Ahmed, M. Arjovsky, V . Dumoulin, and A. Courville, “Improved training of wasserstein gans,” arXiv preprint arXiv:1704.00028, 2017

    2017 arXiv

  14. [22]

    Least squares generative adversarial networks,

    X. Mao, Q. Li, H. Xie, R. Y . Lau, Z. Wang, and S. Paul Smolley, “Least squares generative adversarial networks,” in InternationalConference on Computer Vision, 2017, pp. 2794–2802

    2017

  15. [23]

    Geometric gan,

    J. H. Lim and J. C. Ye, “Geometric gan,” arXiv preprint arXiv:1705.02894, 2017

    2017 arXiv

  16. [24]

    Large scale gan training for high fidelity natural image synthesis,

    A. Brock, J. Donahue, and K. Simonyan, “Large scale gan training for high fidelity natural image synthesis,” arXiv preprint arXiv:1809.11096, 2018

    2018 arXiv

  17. [25]

    Tackling the generative learning trilemma with denoising diffusion gans,

    Z. Xiao, K. Kreis, and A. Vahdat, “Tackling the generative learning trilemma with denoising diffusion gans,” arXiv preprint arXiv:2112.07804, 2021

    2021 arXiv

  18. [26]

    Improved techniques for training gans,

    T. Salimans, I. Goodfellow, W. Zaremba, V . Cheung, A. Radford, and X. Chen, “Improved techniques for training gans,” Conference on Neural Information Processing Systems , vol. 29, pp. 2234–2242, 2016

    2016

  19. [27]

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

    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,” Conference on Neural Information Processing Systems , vol. 30, 2017. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 14

    2017

  20. [28]

    Improved techniques for training score-based generative models,

    Y . Song and S. Ermon, “Improved techniques for training score-based generative models,” Conference on Neural Information Processing Sys- tems, vol. 33, pp. 12 438–12 448, 2020

    2020

  21. [29]

    Deep unsupervised learning using nonequilibrium thermodynamics,

    J. Sohl-Dickstein, E. Weiss, N. Maheswaranathan, and S. Ganguli, “Deep unsupervised learning using nonequilibrium thermodynamics,” in International Conference on Machine Learning . PMLR, 2015, pp. 2256–2265

    2015

  22. [30]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Conference on Neural Information Processing Systems , vol. 33, pp. 6840–6851, 2020

    2020

  23. [31]

    Score-based generative modeling through stochastic differ- ential equations,

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

    2011 arXiv

  24. [32]

    Analytic-dpm: an analytic estimate of the optimal reverse variance in diffusion probabilistic models,

    F. Bao, C. Li, J. Zhu, and B. Zhang, “Analytic-dpm: an analytic estimate of the optimal reverse variance in diffusion probabilistic models,” arXiv preprint arXiv:2201.06503, 2022

    2022 arXiv

  25. [33]

    High- resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” in IEEE/CVF Computer Vision and Pattern Recognition Conference, 2022, pp. 10 684– 10 695

    2022

  26. [34]

    Photorealistic text-to-image diffusion models with deep language understanding,

    C. Saharia, W. Chan, S. Saxena, L. Li, J. Whang, E. L. Denton, K. Ghasemipour, R. Gontijo Lopes, B. Karagol Ayan, T. Salimans et al., “Photorealistic text-to-image diffusion models with deep language understanding,” Conference on Neural Information Processing Systems , vol. 35...

    2022

  27. [35]

    Towards high-quality hdr deghosting with conditional diffusion models,

    Q. Yan, T. Hu, Y . Sun, H. Tang, Y . Zhu, W. Dong, L. Van Gool, and Y . Zhang, “Towards high-quality hdr deghosting with conditional diffusion models,” IEEE Transactions on Circuits and Systems for Video Technology, pp. 1–1, 2023

    2023

  28. [36]

    Adaptive conditional denoising diffusion model with hybrid affinity regularizer for generalized zero-shot learning,

    M. Gao and Q. Dong, “Adaptive conditional denoising diffusion model with hybrid affinity regularizer for generalized zero-shot learning,” IEEE Transactions on Circuits and Systems for Video Technology , pp. 1–1, 2024

    2024

  29. [37]

    Games of gans: Game-theoretical models for generative adversarial networks,

    M. Mohebbi Moghaddam, B. Boroomand, M. Jalali, A. Zareian, A. Daeijavad, M. H. Manshaei, and M. Krunz, “Games of gans: Game-theoretical models for generative adversarial networks,” Artificial Intelligence Review, vol. 56, no. 9, pp. 9771–9807, 2023

    2023

  30. [38]

    Generative adversarial networks as stochastic nash games,

    B. Franci and S. Grammatico, “Generative adversarial networks as stochastic nash games,” arXiv preprint arXiv:2010.10013 , 2020

    2010 arXiv

  31. [39]

    Stackelberg gan: Towards provable minimax equilibrium via multi- generator architectures,

    H. Zhang, S. Xu, J. Jiao, P. Xie, R. Salakhutdinov, and E. P. Xing, “Stackelberg gan: Towards provable minimax equilibrium via multi- generator architectures,” arXiv preprint arXiv:1811.08010 , 2018

    2018 arXiv

  32. [40]

    Finding mixed nash equilibria of generative adversarial networks,

    Y .-P. Hsieh, C. Liu, and V . Cevher, “Finding mixed nash equilibria of generative adversarial networks,” in International Conference on Machine Learning. PMLR, 2019, pp. 2810–2819

    2019

  33. [41]

    Fedgan: Federated gen- erative adversarial networks for distributed data,

    M. Rasouli, T. Sun, and R. Rajagopal, “Fedgan: Federated gen- erative adversarial networks for distributed data,” arXiv preprint arXiv:2006.07228, 2020

    2006 arXiv

  34. [42]

    Dual discriminator generative adversarial nets,

    T. Nguyen, T. Le, H. Vu, and D. Phung, “Dual discriminator generative adversarial nets,” Conference on Neural Information Processing Systems, vol. 30, 2017

    2017

  35. [43]

    Consistency of multiagent distributed generative adversarial networks,

    S. Ke and W. Liu, “Consistency of multiagent distributed generative adversarial networks,” IEEE Transactions on Cybernetics, vol. 52, no. 6, pp. 4886–4896, 2020

    2020

  36. [44]

    Triple generative adversarial nets,

    C. Li, T. Xu, J. Zhu, and B. Zhang, “Triple generative adversarial nets,” Conference on Neural Information Processing Systems , vol. 30, 2017

    2017

  37. [45]

    Many paths to equilibrium: Gans do not need to decrease a divergence at every step,

    W. Fedus, M. Rosca, B. Lakshminarayanan, A. M. Dai, S. Mohamed, and I. Goodfellow, “Many paths to equilibrium: Gans do not need to decrease a divergence at every step,” arXiv preprint arXiv:1710.08446 , 2017

    2017 arXiv

  38. [46]

    Began: Boundary equilib- rium generative adversarial networks,

    D. Berthelot, T. Schumm, and L. Metz, “Began: Boundary equilib- rium generative adversarial networks,” arXiv preprint arXiv:1703.10717, 2017

    2017 arXiv

  39. [47]

    An online learning approach to generative adversarial networks,

    P. Grnarova, K. Y . Levy, A. Lucchi, T. Hofmann, and A. Krause, “An online learning approach to generative adversarial networks,” arXiv preprint arXiv:1706.03269, 2017

    2017 arXiv

  40. [48]

    Fictitious gan: Training gans with historical models,

    H. Ge, Y . Xia, X. Chen, R. Berry, and Y . Wu, “Fictitious gan: Training gans with historical models,” in European Conference on Computer Vision, 2018, pp. 119–134

    2018

  41. [49]

    Image captioning using adversarial networks and reinforcement learning,

    S. Yan, F. Wu, J. S. Smith, W. Lu, and B. Zhang, “Image captioning using adversarial networks and reinforcement learning,” in International Conference on Pattern Recognition . IEEE, 2018, pp. 248–253

    2018

  42. [50]

    Mode regularized gen- erative adversarial networks,

    T. Che, Y . Li, A. P. Jacob, Y . Bengio, and W. Li, “Mode regularized gen- erative adversarial networks,” arXiv preprint arXiv:1612.02136 , 2016

    2016 arXiv

  43. [51]

    Smoothness and stability in gans,

    C. Chu, K. Minami, and K. Fukumizu, “Smoothness and stability in gans,” arXiv preprint arXiv:2002.04185 , 2020

    2002 arXiv

  44. [52]

    Gradient descent gan optimization is locally stable,

    V . Nagarajan and J. Z. Kolter, “Gradient descent gan optimization is locally stable,” arXiv preprint arXiv:1706.04156 , 2017

    2017 arXiv

  45. [53]

    Stabilizing train- ing of generative adversarial networks through regularization,

    K. Roth, A. Lucchi, S. Nowozin, and T. Hofmann, “Stabilizing train- ing of generative adversarial networks through regularization,” arXiv preprint arXiv:1705.09367, 2017

    2017 arXiv

  46. [54]

    Which training methods for gans do actually converge?

    L. Mescheder, A. Geiger, and S. Nowozin, “Which training methods for gans do actually converge?” in International Conference on Machine Learning. PMLR, 2018, pp. 3481–3490

    2018

  47. [55]

    Wasserstein generative ad- versarial networks,

    M. Arjovsky, S. Chintala, and L. Bottou, “Wasserstein generative ad- versarial networks,” in International Conference on Machine Learning . PMLR, 2017, pp. 214–223

    2017

  48. [56]

    Hierarchical implicit models and likelihood-free variational inference,

    D. Tran, R. Ranganath, and D. M. Blei, “Hierarchical implicit models and likelihood-free variational inference,” vol. 2017-December, 2017

    2017

  49. [57]

    f-gan: Training generative neu- ral samplers using variational divergence minimization,

    S. Nowozin, B. Cseke, and R. Tomioka, “f-gan: Training generative neu- ral samplers using variational divergence minimization,” in Conference on Neural Information Processing Systems , 2016, pp. 271–279

    2016

  50. [58]

    Varia- tional inference via wasserstein gradient flows,

    M. Lambert, S. Chewi, F. Bach, S. Bonnabel, and P. Rigollet, “Varia- tional inference via wasserstein gradient flows,” Conference on Neural Information Processing Systems , vol. 35, pp. 14 434–14 447, 2022

    2022

  51. [59]

    Variational wasserstein gradient flow,

    J. Fan, Q. Zhang, A. Taghvaei, and Y . Chen, “Variational wasserstein gradient flow,” arXiv preprint arXiv:2112.02424 , 2021

    2021 arXiv

  52. [60]

    Deep gener- ative learning via variational gradient flow,

    Y . Gao, Y . Jiao, Y . Wang, Y . Wang, C. Yang, and S. Zhang, “Deep gener- ative learning via variational gradient flow,” in International Conference on Machine Learning . PMLR, 2019, pp. 2093–2101

    2019

  53. [61]

    Gradient layer: Enhancing the convergence of adversarial training for generative models,

    A. Nitanda and T. Suzuki, “Gradient layer: Enhancing the convergence of adversarial training for generative models,” in International Confer- ence on Artificial Intelligence and Statistics . PMLR, 2018, pp. 1008– 1016

    2018

  54. [62]

    How well generative adversarial networks learn distributions,

    T. Liang, “How well generative adversarial networks learn distributions,” Journal of Machine Learning Research, vol. 22, no. 228, pp. 1–41, 2021

    2021

  55. [63]

    An error analysis of generative adversarial networks for learning distributions,

    J. Huang, Y . Jiao, Z. Li, S. Liu, Y . Wang, and Y . Yang, “An error analysis of generative adversarial networks for learning distributions,” Journal of Machine Learning Research , vol. 23, no. 116, pp. 1–43, 2022

    2022

  56. [64]

    Error analysis of generative adversarial net- work,

    M. Hasan and H. Sang, “Error analysis of generative adversarial net- work,” arXiv preprint arXiv:2310.15387 , 2023

    2023 arXiv

  57. [65]

    Training generative adversarial networks in one stage,

    C. Shen, Y . Yin, X. Wang, X. Li, J. Song, and M. Song, “Training generative adversarial networks in one stage,” in IEEE/CVF Computer Vision and Pattern Recognition Conference , 2021, pp. 3350–3360

    2021

  58. [66]

    The numerics of gans,

    L. Mescheder, S. Nowozin, and A. Geiger, “The numerics of gans,” arXiv preprint arXiv:1705.10461 , 2017. Chang Wan Chang Wan received a Bachelor’s and Master’s degree from Nanchang University, China, in 2009 and 2013, respectively. Now, he is a Ph.D. student at Zhejiang Normal...

    2017 arXiv

  59. [2006]

    His research interests include intelligent information processing and geographic information systems

    He is currently a Professor with the School of Computer Science and Technology, Zhejiang Nor- mal University, Jinhua, China, and also with the School of Information Engineering, Huzhou Univer- sity, Huzhou, China. His research interests include intelligent information processi...

    1999

Pith tools

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