The reviewed record of science sign in
Pith

arxiv: 2606.27246 · v1 · pith:Q5EU2T5S · submitted 2026-06-25 · cs.LG

Effective Covariance Dynamics in Solvable High-Dimensional GANs

Andrew Bond , Zafer Do\u{g}an This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-26 04:45 UTCgrok-4.3pith:Q5EU2T5Srecord.jsonopen to challenge →

classification cs.LG
keywords GAN training dynamicshigh-dimensional limiteffective covariancestability analysislatent structurequadratic discriminatorsolvable modelsubspace recovery
0
0 comments X

The pith

In high-dimensional GANs, stochastic training converges to deterministic ODEs governed by a probability-weighted effective covariance that sets mode-wise learnability intervals.

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

The paper shows that when a linear generator learns a low-dimensional subspace from data with class-dependent or correlated latent structure, all heterogeneity collapses into one effective second moment for a quadratic energy discriminator. This reduction lets the microscopic stochastic process converge to deterministic ODEs in the high-dimensional limit. Stability analysis then produces an explicit interval for each effective mode: learning starts once the leading eigenvalue exceeds a lower threshold set by learning rates and noise, and full recovery requires every relevant mode to stay inside the interval. Low-rank correlations can therefore push weak directions above the threshold while overly strong ones push modes out and destabilize recovery. Simulations match the ODE trajectories and phase boundaries, and experiments on MNIST, FashionMNIST, and CIFAR-10 confirm that supplying the generator with an informed covariance improves alignment with the data subspace.

Core claim

For the quadratic energy discriminator, all latent heterogeneity enters the dynamics through a probability-weighted effective second moment. The stochastic microscopic training process converges, in the high-dimensional limit, to deterministic ordinary differential equations governed by this effective covariance. In the matched-covariance specialization, the stability analysis yields a mode-wise solvable interval determined by the learning rates and noise level: learning begins when the leading effective eigenvalue crosses the lower threshold, while full recovery requires all relevant effective modes to remain within the interval. This reveals a signal-boosting mechanism: low-rank correlatio

What carries the argument

The probability-weighted effective second moment (effective covariance) that folds all class-dependent, correlated, and non-zero-mean latent structure into the quadratic discriminator dynamics.

If this is right

  • The high-dimensional training trajectory is fully determined by the eigenvalues of the effective covariance and the chosen learning rates and noise.
  • Low-rank correlations in the latent structure can raise weak modes above the lower threshold and enable their recovery.
  • Correlations that are too strong push modes outside the solvable interval and prevent full subspace recovery.
  • Supplying the generator with a covariance that matches the data-driven reference improves alignment on image datasets.

Where Pith is reading between the lines

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

  • The same reduction might be attempted for non-quadratic discriminators if an analogous effective moment can be derived.
  • The mode-wise interval suggests a practical test: monitor effective eigenvalues during training to predict which directions will converge.
  • The boosting mechanism could be used to design latent priors that deliberately lift weak features in other generative models.

Load-bearing premise

All latent heterogeneity reduces to a single effective second moment only when the discriminator is quadratic and the system is taken in the high-dimensional limit.

What would settle it

Numerical simulations of the finite-dimensional stochastic GAN updates that fail to approach the predicted ODE trajectories or violate the calculated phase boundaries as dimension grows.

read the original abstract

We study a solvable high-dimensional model of generative adversarial network (GAN) training in which a linear generator learns a low-dimensional subspace from data with structured latent covariance. Prior solvable GAN analyses assume unconditional signals with diagonal latent covariance; we extend the multi-feature discriminator setting to class-dependent, correlated, and non-zero-mean latent structure. For the quadratic energy discriminator, all such heterogeneity enters the dynamics through a probability-weighted effective second moment. We prove that the stochastic microscopic training process converges, in the high-dimensional limit, to deterministic ordinary differential equations governed by this effective covariance. In the matched-covariance specialization, the stability analysis yields a mode-wise solvable interval determined by the learning rates and noise level: learning begins when the leading effective eigenvalue crosses the lower threshold, while full recovery requires all relevant effective modes to remain within the interval. This reveals a signal-boosting mechanism: low-rank correlations can lift weak directions above the learnability threshold, whereas overly strong correlations destabilize recovery. Numerical simulations validate the ODE, phase boundary, and boosting mechanism. Experiments on MNIST, FashionMNIST, and CIFAR-10 further show that informed generator covariance improves alignment with the data-driven reference subspace.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

0 major / 2 minor

Summary. The manuscript develops a solvable high-dimensional model of GAN training in which a linear generator learns a low-dimensional subspace from data whose latent structure may include class dependence, correlations, and non-zero means. For the quadratic energy discriminator, all such heterogeneity reduces to dynamics governed by a probability-weighted effective second-moment matrix. The paper proves that the stochastic microscopic training dynamics converge to deterministic ODEs in the high-dimensional limit, derives a mode-wise stability interval in the matched-covariance specialization (determined by learning rates and noise level), and identifies a signal-boosting mechanism whereby low-rank correlations can lift weak modes above the learnability threshold while overly strong correlations destabilize recovery. Numerical simulations and experiments on MNIST, FashionMNIST, and CIFAR-10 are presented in support.

Significance. If the stated convergence and stability results hold, the work supplies a rare analytically tractable window into how structured latent covariances affect GAN learnability, including an explicit boosting mechanism. The scoping to the quadratic discriminator and the high-dimensional limit is clearly stated, and the stability thresholds are expressed directly in terms of externally chosen parameters (learning rates, noise) rather than quantities fitted from the target data, so the circularity concern does not apply. The combination of rigorous reduction to effective covariance, mode-wise solvability, and empirical validation on real data constitutes a solid contribution to the theory of high-dimensional generative models.

minor comments (2)
  1. [Abstract / §3] The abstract asserts proofs of convergence to ODEs and mode-wise stability intervals; the manuscript should ensure that the main text explicitly flags the precise high-dimensional scaling assumptions and any error bounds or rates that accompany the limit statement.
  2. [§2] Notation for the effective covariance matrix and the probability weights should be introduced with a single consolidated definition early in the paper to avoid repeated cross-references when the matched-covariance specialization is introduced.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of the manuscript, the clear summary of its contributions, and the recommendation of minor revision. No specific major comments were raised in the report.

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper derives the effective covariance reduction and convergence to deterministic ODEs as a high-dimensional limit proof specific to the quadratic energy discriminator, with stability thresholds expressed directly in terms of externally specified learning rates and noise levels. No steps reduce by construction to fitted inputs from the target data, self-citations bearing the central claim, or ansatzes smuggled via prior work. Numerical simulations and MNIST/FashionMNIST/CIFAR-10 experiments provide independent validation outside the derivation. This matches the default expectation of a non-circular analysis.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on the high-dimensional limit and the quadratic discriminator assumption; no free parameters are explicitly fitted in the abstract description, and no new entities are postulated.

axioms (2)
  • domain assumption High-dimensional limit in which the stochastic microscopic training process converges to deterministic ODEs
    Invoked to obtain the effective covariance dynamics from the microscopic process.
  • domain assumption Quadratic energy discriminator such that all latent heterogeneity enters only through the probability-weighted effective second moment
    Required for the reduction that makes the model solvable.

pith-pipeline@v0.9.1-grok · 5733 in / 1525 out tokens · 38172 ms · 2026-06-26T04:45:20.259403+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

79 extracted references · 5 canonical work pages · 2 internal anchors

  1. [1]

    Advances in Neural Information Processing Systems , year=

    A Solvable High-Dimensional Model of GAN , author=. Advances in Neural Information Processing Systems , year=

  2. [2]

    Schubert Varieties and Distances between Subspaces of Different Dimensions , author=. SIAM J. Matrix Anal. Appl. , year=

  3. [3]

    International Conference on Machine Learning , year=

    Stabilizing GANs' Training with Brownian Motion Controller , author=. International Conference on Machine Learning , year=

  4. [4]

    Conference on Uncertainty in Artificial Intelligence , year=

    Towards a Better Understanding and Regularization of GAN Training Dynamics , author=. Conference on Uncertainty in Artificial Intelligence , year=

  5. [5]

    On Finding Local Nash Equilibria (and Only Local Nash Equilibria) in Zero-Sum Games , author=

  6. [6]

    Advances in Neural Information Processing Systems , year=

    Time-series Generative Adversarial Networks , author=. Advances in Neural Information Processing Systems , year=

  7. [7]

    International Conference on Natural Language Generation , year=

    GAN-LM: Generative Adversarial Network using Language Models for Downstream Applications , author=. International Conference on Natural Language Generation , year=

  8. [8]

    2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    A Style-Based Generator Architecture for Generative Adversarial Networks , author=. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    2019

  9. [9]

    2021 IEEE/CVF International Conference on Computer Vision (ICCV) , year=

    StyleCLIP: Text-Driven Manipulation of StyleGAN Imagery , author=. 2021 IEEE/CVF International Conference on Computer Vision (ICCV) , year=

    2021

  10. [10]

    Communications of the ACM , year=

    Generative adversarial networks , author=. Communications of the ACM , year=

  11. [11]

    IEEE Journal of Selected Topics in Signal Processing , year=

    Subspace Estimation From Incomplete Observations: A High-Dimensional Analysis , author=. IEEE Journal of Selected Topics in Signal Processing , year=

  12. [12]

    2016 IEEE Information Theory Workshop (ITW) , year=

    Online learning for sparse PCA in high dimensions: Exact dynamics and phase transitions , author=. 2016 IEEE Information Theory Workshop (ITW) , year=

    2016

  13. [13]

    IEEE Trans

    Projection approximation subspace tracking , author=. IEEE Trans. Signal Process. , year=

  14. [14]

    2010 48th Annual Allerton Conference on Communication, Control, and Computing (Allerton) , year=

    Online identification and tracking of subspaces from highly incomplete information , author=. 2010 48th Annual Allerton Conference on Communication, Control, and Computing (Allerton) , year=

    2010

  15. [15]

    Journal of Mathematical Biology , year=

    Simplified neuron model as a principal component analyzer , author=. Journal of Mathematical Biology , year=

  16. [16]

    IEEE Signal Processing Magazine , volume=

    The mnist database of handwritten digit images for machine learning research , author=. IEEE Signal Processing Magazine , volume=. 2012 , publisher=

    2012

  17. [17]

    Proceedings of the IEEE , year=

    Streaming PCA and Subspace Tracking: The Missing Data Case , author=. Proceedings of the IEEE , year=

  18. [18]

    International Conference on Artificial Intelligence and Statistics , year=

    On the equivalence of Oja's algorithm and GROUSE , author=. International Conference on Artificial Intelligence and Statistics , year=

  19. [19]

    International Conference on Artificial Intelligence and Statistics , year=

    Learning Dynamics in Linear VAE: Posterior Collapse Threshold, Superfluous Latent Space Pitfalls, and Speedup with KL Annealing , author=. International Conference on Artificial Intelligence and Statistics , year=

  20. [20]

    Journal of Statistical Mechanics: Theory and Experiment , year=

    Phase diagram of stochastic gradient descent in high-dimensional two-layer neural networks , author=. Journal of Statistical Mechanics: Theory and Experiment , year=

  21. [21]

    Journal of Statistical Mechanics: Theory and Experiment , year=

    The dynamics of representation learning in shallow, non-linear autoencoders , author=. Journal of Statistical Mechanics: Theory and Experiment , year=

  22. [22]

    Proceedings of Machine Learning Research , year=

    From high-dimensional & mean-field dynamics to dimensionless ODEs: A unifying approach to SGD in two-layers networks , author=. Proceedings of Machine Learning Research , year=

  23. [23]

    AT&T Laboratories Cambridge , title=

  24. [24]

    Journal of the American Statistical Association , year=

    On Consistency and Sparsity for Principal Components Analysis in High Dimensions , author=. Journal of the American Statistical Association , year=

  25. [25]

    International Conference on Learning Representations , year=

    Energy-based Generative Adversarial Networks , author=. International Conference on Learning Representations , year=

  26. [26]

    , year =

    Alex Krizhevsky and Vinod Nair and Geoffrey Hinton , title =. , year =

  27. [27]

    Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms

    Fashion-MNIST: a novel image dataset for benchmarking machine learning algorithms , author=. arXiv preprint arXiv:1708.07747 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  28. [28]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages=

    The unreasonable effectiveness of deep features as a perceptual metric , author=. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages=

  29. [29]

    Advances in Neural Information Processing Systems , volume=

    Exploring the precise dynamics of single-layer GAN models: Leveraging multi-feature discriminators for high-dimensional subspace learning , author=. Advances in Neural Information Processing Systems , volume=

  30. [30]

    2012 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    PETRELS: Subspace estimation and tracking from partial observations , author=. 2012 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2012 , organization=

    2012

  31. [31]

    Forty-second International Conference on Machine Learning , year=

    Neural Collapse Beyond the Unconstrained Features Model: Landscape, Dynamics, and Generalization in the Mean-Field Regime , author=. Forty-second International Conference on Machine Learning , year=

  32. [32]

    2019 57th Annual Allerton Conference on Communication, Control, and Computing (Allerton) , pages=

    Wasserstein gan can perform pca , author=. 2019 57th Annual Allerton Conference on Communication, Control, and Computing (Allerton) , pages=

    2019

  33. [33]

    Advances in Neural Information Processing Systems , volume=

    Gans trained by a two time-scale update rule converge to a local nash equilibrium , author=. Advances in Neural Information Processing Systems , volume=

  34. [34]

    The 28th International Conference on Artificial Intelligence and Statistics , year=

    Understanding the Learning Dynamics of LoRA: A Gradient Flow Perspective on Low-Rank Adaptation in Matrix Factorization , author=. The 28th International Conference on Artificial Intelligence and Statistics , year=

  35. [35]

    Forty-second International Conference on Machine Learning , year=

    Propagation of Chaos for Mean-Field Langevin Dynamics and its Application to Model Ensemble , author=. Forty-second International Conference on Machine Learning , year=

  36. [36]

    International Conference on Learning Representations , year=

    Improving Probabilistic Diffusion Models With Optimal Diagonal Covariance Matching , author=. International Conference on Learning Representations , year=

  37. [37]

    Journal of mathematical analysis and applications , volume=

    On stochastic approximation of the eigenvectors and eigenvalues of the expectation of a random matrix , author=. Journal of mathematical analysis and applications , volume=. 1985 , publisher=

    1985

  38. [38]

    Implicitly Normalized Online PCA: A Regularized Algorithm with Exact High-Dimensional Dynamics , author=

  39. [39]

    Neural Information Processing Systems , year=

    Robust Streaming PCA , author=. Neural Information Processing Systems , year=

  40. [40]

    Nature Reviews Methods Primers , volume=

    Principal component analysis , author=. Nature Reviews Methods Primers , volume=. 2022 , publisher=

    2022

  41. [41]

    Proceedings of the 34th International Conference on Machine Learning , series =

    Wasserstein Generative Adversarial Networks , author =. Proceedings of the 34th International Conference on Machine Learning , series =. 2017 , publisher =

    2017

  42. [42]

    Advances in Neural Information Processing Systems , volume =

    Improved Training of Wasserstein GANs , author =. Advances in Neural Information Processing Systems , volume =

  43. [43]

    International conference on machine learning , pages=

    Which training methods for GANs do actually converge? , author=. International conference on machine learning , pages=. 2018 , organization=

    2018

  44. [44]

    International Conference on Learning Representations , year =

    Unrolled Generative Adversarial Networks , author =. International Conference on Learning Representations , year =

  45. [45]

    International Conference on Learning Representations , year =

    Training GANs with Optimism , author =. International Conference on Learning Representations , year =

  46. [46]

    Advances in Neural Information Processing Systems , volume =

    Gradient Descent GAN Optimization is Locally Stable , author =. Advances in Neural Information Processing Systems , volume =

  47. [47]

    Conditional Generative Adversarial Nets

    Conditional Generative Adversarial Nets , author =. arXiv preprint arXiv:1411.1784 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  48. [48]

    international conference on machine learning , volume =

    Conditional Image Synthesis with Auxiliary Classifier GANs , author =. international conference on machine learning , volume =. 2017 , publisher =

    2017

  49. [49]

    Advances in Neural Information Processing Systems , volume =

    InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets , author =. Advances in Neural Information Processing Systems , volume =

  50. [50]

    International Conference on Learning Representations Workshop , year =

    -VAE: Learning Basic Visual Concepts with a Constrained Variational Framework , author =. International Conference on Learning Representations Workshop , year =

  51. [51]

    international conference on machine learning , pages=

    Challenging common assumptions in the unsupervised learning of disentangled representations , author=. international conference on machine learning , pages=. 2019 , organization=

    2019

  52. [52]

    The Annals of Probability , volume =

    Phase Transition of the Largest Eigenvalue for Nonnull Complex Sample Covariance Matrices , author =. The Annals of Probability , volume =. 2005 , doi =

    2005

  53. [53]

    Advances in Mathematics , volume=

    The eigenvalues and eigenvectors of finite, low rank perturbations of large random matrices , author=. Advances in Mathematics , volume=. 2011 , publisher=

    2011

  54. [54]

    Advances in Neural Information Processing Systems , volume =

    Ganspace: Discovering interpretable gan controls , author =. Advances in Neural Information Processing Systems , volume =

  55. [55]

    In: Proceedings of the 34th International Conference on Machine Learning, Proceedings of Machine Learning Research, vol 70

    Arjovsky M, Chintala S, Bottou L (2017) Wasserstein generative adversarial networks. In: Proceedings of the 34th International Conference on Machine Learning, Proceedings of Machine Learning Research, vol 70. PMLR, pp 214--223

    2017

  56. [56]

    Phase transition of the largest eigenvalue for nonnull complex sample covariance matri- ces,

    Baik J, Ben Arous G, P \'e ch \'e S (2005) Phase transition of the largest eigenvalue for nonnull complex sample covariance matrices. The Annals of Probability 33(5):1643--1697. doi:10.1214/009117905000000233

    work page doi:10.1214/009117905000000233 2005

  57. [57]

    Advances in Mathematics 227(1):494--521

    Benaych-Georges F, Nadakuditi RR (2011) The eigenvalues and eigenvectors of finite, low rank perturbations of large random matrices. Advances in Mathematics 227(1):494--521

    2011

  58. [58]

    Advances in Neural Information Processing Systems 37:69865--69887

    Bond A, Do g an Z (2024) Exploring the precise dynamics of single-layer gan models: Leveraging multi-feature discriminators for high-dimensional subspace learning. Advances in Neural Information Processing Systems 37:69865--69887

    2024

  59. [59]

    In: Advances in Neural Information Processing Systems

    Chen X, Duan Y, Houthooft R, et al (2016) Infogan: Interpretable representation learning by information maximizing generative adversarial nets. In: Advances in Neural Information Processing Systems

    2016

  60. [60]

    In: International Conference on Learning Representations

    Daskalakis C, Ilyas A, Syrgkanis V, et al (2018) Training gans with optimism. In: International Conference on Learning Representations

    2018

  61. [61]

    IEEE Signal Processing Magazine 29(6):141--142

    Deng L (2012) The mnist database of handwritten digit images for machine learning research. IEEE Signal Processing Magazine 29(6):141--142

    2012

  62. [62]

    Generative adversarial networks,

    Goodfellow IJ, Pouget-Abadie J, Mirza M, et al (2014) Generative adversarial networks. Communications of the ACM 63:139 -- 144. doi:10.1145/3422622

    work page doi:10.1145/3422622 2014

  63. [63]

    In: Advances in Neural Information Processing Systems

    Gulrajani I, Ahmed F, Arjovsky M, et al (2017) Improved training of wasserstein gans. In: Advances in Neural Information Processing Systems

    2017

  64. [64]

    Advances in Neural Information Processing Systems 30

    Heusel M, Ramsauer H, Unterthiner T, et al (2017) Gans trained by a two time-scale update rule converge to a local nash equilibrium. Advances in Neural Information Processing Systems 30

    2017

  65. [65]

    In: International Conference on Learning Representations Workshop

    Higgins I, Matthey L, Pal A, et al (2017) -vae: Learning basic visual concepts with a constrained variational framework. In: International Conference on Learning Representations Workshop

    2017

  66. [66]

    In: Advances in Neural Information Processing Systems

    Härkönen E, Hertzmann A, Lehtinen J, et al (2020) Ganspace: Discovering interpretable gan controls. In: Advances in Neural Information Processing Systems

    2020

  67. [67]

    Journal of the American Statistical Association 104:682 -- 693

    Johnstone IM, Lu AY (2009) On consistency and sparsity for principal components analysis in high dimensions. Journal of the American Statistical Association 104:682 -- 693

    2009

  68. [68]

    2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) pp 4401--4410

    Karras T, Laine S, Aila T (2019) A style-based generator architecture for generative adversarial networks. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) pp 4401--4410

    2019

  69. [69]

    Krizhevsky A, Nair V, Hinton G (2009) Cifar-10 (canadian institute for advanced research)

    2009

  70. [70]

    In: international conference on machine learning, PMLR, pp 4114--4124

    Locatello F, Bauer S, Lucic M, et al (2019) Challenging common assumptions in the unsupervised learning of disentangled representations. In: international conference on machine learning, PMLR, pp 4114--4124

    2019

  71. [71]

    Mescheder L, Geiger A, Nowozin S (2018) Which training methods for gans do actually converge? In: International conference on machine learning, PMLR, pp 3481--3490

    2018

  72. [72]

    In: International Conference on Learning Representations

    Metz L, Poole B, Pfau D, et al (2017) Unrolled generative adversarial networks. In: International Conference on Learning Representations

    2017

  73. [73]

    arXiv preprint arXiv:14111784

    Mirza M, Osindero S (2014) Conditional generative adversarial nets. arXiv preprint arXiv:14111784

    2014

  74. [74]

    In: Advances in Neural Information Processing Systems

    Nagarajan V, Kolter JZ (2017) Gradient descent gan optimization is locally stable. In: Advances in Neural Information Processing Systems

    2017

  75. [75]

    In: international conference on machine learning, vol 70

    Odena A, Olah C, Shlens J (2017) Conditional image synthesis with auxiliary classifier gans. In: international conference on machine learning, vol 70. PMLR, pp 2642--2651

    2017

  76. [76]

    IEEE Journal of Selected Topics in Signal Processing 12:1240--1252

    Wang C, Eldar YC, Lu YM (2018 a ) Subspace estimation from incomplete observations: A high-dimensional analysis. IEEE Journal of Selected Topics in Signal Processing 12:1240--1252. doi:10.1109/jstsp.2018.2877405

    work page doi:10.1109/jstsp.2018.2877405 2018

  77. [77]

    In: Advances in Neural Information Processing Systems

    Wang C, Hu H, Lu YM (2018 b ) A solvable high-dimensional model of gan. In: Advances in Neural Information Processing Systems

    2018

  78. [78]

    arXiv preprint arXiv:170807747

    Xiao H, Rasul K, Vollgraf R (2017) Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms. arXiv preprint arXiv:170807747

    2017

  79. [79]

    In: International Conference on Learning Representations

    Zhao J, Mathieu M, LeCun Y (2017) Energy-based generative adversarial networks. In: International Conference on Learning Representations

    2017