Pith. sign in

REVIEW 4 major objections 5 minor 73 references

AutoGAN: Neural Architecture Search for Generative Adversarial Networks

T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read AutoGAN extends neural architecture search to GANs, searching only the generator and reporting state-of-the-art FID scores of 12.42 on CIFAR-10 and 31.01 on STL-10.

desk verdict First credible NAS-for-GANs paper; results are solid but the SOTA claim rests on an underspecified discriminator-growth routine that peer review should pin down. read the letter →

arxiv 1908.03835 v1 pith:2JI2JDEY submitted 2019-08-11 cs.CV cs.LGeess.IV

classification cs.CVcs.LGeess.IV
keywords AutoGANneuralarchitecturesearchgenerativeadversarialnetworksgeneratorInceptionscorerewardFréchetdistancemulti-levelparametersharing
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that the architecture of a GAN generator can be discovered by neural architecture search rather than designed by hand, and that the discovered generator beats hand-crafted ones. It introduces AutoGAN, in which an RNN controller selects each cell's convolution block, normalization, upsampling type, and skip connections while the discriminator follows a fixed growth routine. Using Inception score as the reward, parameter sharing, dynamic resetting, and progressive cell-by-cell search, AutoGAN reports a CIFAR-10 FID of 12.42 and an STL-10 FID of 31.01, both state-of-the-art at publication, with competitive Inception scores. If correct, this opens GAN generator design to automation the way classification architectures were automated earlier.

What carries the argument

The load-bearing object is the per-cell search-space tuple $(skip_1,\dots,skip_s,C,N,U,SC)$, where $skip_i$ enables a connection from cell $i-1$, $C$ is a pre- or post-activation convolution block, $N$ is batch norm, instance norm, or none, $U$ is bilinear, nearest-neighbor, or deconvolution upsampling, and $SC$ is an in-cell shortcut. An LSTM controller emits these choices autoregressively; its hidden state is reused in the next search stage, and top-$K$ beams continue to the next cell. Parameter sharing lets many sampled child generators share weights during search, dynamic resetting reinitializes shared GAN parameters when loss standard deviation falls below a threshold, and multi-level search grows cells bottom-up. The reward for each sampled architecture is the Inception score, computed cheaply on the shared model.

What would settle it

Retrain the discovered CIFAR-10 generator and the strongest hand-crafted baselines from scratch under matched training budgets, seeds, and hyperparameters; if the FID gap to 12.42 does not reproduce or falls within run-to-run noise, the search advantage is not established. Alternatively, rerun the AutoGAN search with a different fixed discriminator growth routine; if FID worsens substantially, the discriminator schedule is doing the work.

Watch

Extended reading notes

Core claim

AutoGAN's central discovery is that a search algorithm can rival, and on FID surpass, hand-crafted generator architectures in unconditional image generation. The controller builds generators cell by cell from a search space of block type, normalization, upsampling, in-cell shortcut, and skip connections to earlier cells; it is trained by REINFORCE with Inception score as reward. A parameter dynamic-resetting strategy restarts the shared GAN when training-loss variance collapses, and a multi-level strategy searches cells progressively with beam search. On CIFAR-10 the found generator reaches Inception score 8.55 and FID 12.42, improving on prior FID results. The same architecture retrained on STL-10 gives FID 31.01, also state-of-the-art, and the paper's ablations show its proxy correlates with true Inception score (rank correlation 0.779), that FID as reward gives comparable results, and that multi-level search and dynamic resetting each help.

Load-bearing premise

The central assumption is that the fixed discriminator-growth routine keeps the generator/discriminator capacity balance intact; if that routine is unbalanced or ad hoc, the reported gains may be due to the discriminator schedule rather than to the searched generator, and the paper leaves the routine's details to the supplementary.

Editorial extensions

If this is right

  • A discovered generator architecture, not a new loss or training trick, is enough to set new state-of-the-art FID on CIFAR-10 and STL-10.
  • GAN generator search is feasible without searching the discriminator: fixing D to a growing routine with spectral normalization keeps training stable enough for reinforcement-learning-based NAS.
  • The CIFAR-10-found architecture transfers to STL-10, suggesting searched GAN generators can generalize across datasets and resolutions.
  • An Inception-score proxy with parameter sharing is a workable reward: proxy and true Inception scores correlate with rank correlation 0.779, and searching with FID as reward reaches comparable quality.
  • Multi-level progressive search outperforms single-level search both in final Inception score and in training time.

Reading between the lines

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

  • An implicit test the paper does not run: replace the fixed discriminator growth routine with a different one and re-run the search; equal FID would strengthen the generator-search claim, while a drop would suggest the discriminator schedule does much of the work.
  • Because the search space includes SN-GAN but excludes attention, style-based, and multi-discriminator generators, the comparison is to a subset of hand-crafted GANs; searching with those blocks as candidates would test whether the search advantage persists outside the covered space.
  • The reported 43-hour CIFAR-10 search cost suggests a direct extension: search at lower resolution or with more aggressive weight sharing, then fine-tune or transfer the discovered generator to higher resolution, mirroring what classification NAS has done.
  • The discovered preferences—pre-activation blocks, nearest or bilinear upsampling over deconvolution, no normalization, and dense skip connections—could be used as a prior to shrink the search space in future work, but the paper does not claim this.
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. AutoGAN introduces the first neural architecture search framework for generative adversarial networks. The authors define a generator cell search space over block type, normalization, upsampling, skip connections, and in-cell shortcuts, and train an RNN controller by REINFORCE using Inception Score as the reward. The search is accelerated by ENAS-style parameter sharing, a dynamic-resetting strategy for collapsed shared models, and a multi-level progressive architecture search with beam search. On CIFAR-10, the discovered generator reaches IS 8.55 and FID 12.42, and the same architecture transferred to STL-10 reaches IS 9.16 and FID 31.01. The paper also provides ablations comparing IS versus FID reward, dynamic resetting, MLAS versus SLAS, and random search baselines, and closes with a discussion of search-space, resolution, and discriminator-search limitations.

Significance. If the results hold, the paper is significant as a first proof-of-concept for architecture search in GANs and provides evidence that generator macro-architecture affects FID. The strengths include a detailed method description, a quantitative proxy-reward correlation (Spearman 0.779), an explicit comparison between IS-based and FID-based rewards, a speedup analysis for dynamic resetting, a comparison to random search, and a public code link. The reported FID improvements over hand-crafted baselines are potentially important. However, the central state-of-the-art claim currently rests on an underspecified discriminator growth routine and on a selection protocol that is not fully controlled; these issues should be addressed before the claim can be taken at face value.

major comments (4)
  1. [Section 3, 'Technical Approach' (first paragraph)] The discriminator growth routine is load-bearing and underspecified. The paper states that D grows "by following a given routine to stack pre-defined blocks" and defers the details to a supplementary file; the arXiv v1 contains no supplementary, and no pinned commit of the released code is identified. Because the paper explicitly rejects both fixed-D search and joint G/D search on capacity-imbalance grounds, this D schedule is the mechanism that is supposed to keep the shared GAN trainable during the search. Without a precise specification of the routine, or an ablation or sensitivity study over different D growth choices, the reported FID gains cannot be attributed to the NAS search of G as opposed to the hand-designed discriminator schedule. Please provide the full routine, a citation if it is a standard one, and a sensitivity check.
  2. [Section 3.3.3, 'Architecture Derivation'; Table 1] The final architecture is selected after retraining the top-K candidates by highest Inception Score, so the reported IS of 8.55 is a selected maximum and the IS comparison in Table 1 is partly by construction. The corresponding FID values of the top-1, top-2, and top-3 candidates differ (12.42, 13.67, 13.87), but no repeated-search or retraining-seed intervals are provided for any FID number. To support the state-of-the-art FID claim, the paper should report the distribution of FID over search repetitions or at least over final retraining seeds, and should report FID for all derived candidates rather than only for the IS-selected architecture.
  3. [Section 4.1, Table 1; Section 4.2, Table 2] The comparison baselines are collected from their original papers and may therefore use different FID evaluation protocols. The manuscript does not state the number of generated samples used for FID computation, the reference implementation/version, or any confidence intervals for FID. Since the central claim is a new state-of-the-art FID, the evaluation protocol should be stated precisely and, ideally, the baselines should be recomputed under the same protocol. At minimum, the FID variance over seeds must be reported.
  4. [Section 4.2, Table 2] The STL-10 result uses the CIFAR-10-discovered architecture without re-searching on STL-10; it is an impressive transfer result but not an STL-10 search result. The abstract's phrasing "setting new state-of-the-art FID scores ... 31.01 on STL-10" should be qualified as "with a transferred architecture," and the comparison to methods whose architectures may have been tuned on STL-10 should be framed as a transferability study rather than a direct search comparison.
minor comments (5)
  1. [Abstract and Section 4.3.1] There are several typos: "avaliable" in the abstract, "CFIAR-10" in Section 4.3.1, "unlabled" in Section 4.2, and "performace" in the Figure 9 caption.
  2. [Section 4.3.2, Figure 8] The claim that IS-driven and FID-driven searches show "comparable performance" is supported only by plots; please report the quantitative final IS and FID values for both searches so the reader can judge the size of the difference.
  3. [Section 4.3.4, Figure 10] The MLAS versus SLAS ablation reports only Inception Score; since the paper's headline metric is FID, please report FID for both search strategies as well.
  4. [Section 1 and Section 4.1] The statement that the search space "can only cover SN-GAN" is not fully explained; since spectral normalization is applied only to the discriminator in this paper, please clarify why SN-GAN's generator is covered by the search space while the other compared methods are not.
  5. [Algorithm 1] The pseudo-code's signature "train(generator,discriminator,FDR)" is slightly ambiguous because FDR is both an input and an output of the training procedure; please make the data flow of the dynamic-resetting flag explicit.

Circularity Check

1 steps flagged · score 4.0 of 10

The reported Inception score is selection-on-the-reward and partially fitted; the FID state-of-the-art claim is external and non-circular.

  1. fitted input called prediction [Section 3.3.3 (Architecture Derivation); reward defined in Section 3.2 and Abstract]
    "We will first sample several generator architectures from the learned policy π(a,θ). Then, the reward R (Inception score) will be calculated for each model. We will then pick topK models in terms of highest rewards, and train them from scratch. After that, we evaluate their Inception scores again, and the model with the highest Inception score becomes our final derived generator architecture."

    Section 3.2 sets the controller reward to IS ('we choose the IS of each derived child model, as the reward'), and Section 3.3.3 selects the final architecture as the candidate with the highest IS after training from scratch. The reported headline IS 8.55 is therefore the maximum over IS-evaluated candidates, i.e., the optimized objective itself rather than an independent estimate; selecting on the same metric statistically forces the number upward. The FID 12.42 is not optimized, and the paper's FID-reward ablation shows comparable behavior, so the central SOTA claim remains external to the search objective. The IS result is a fitted/selected quantity, not an independent confirmation.

full rationale

The paper's central contribution is the NAS search for GAN generators, and its strongest evidence is the FID score of 12.42 on CIFAR-10 and 31.01 on STL-10. FID is not used as the controller reward; the search optimizes IS, and the paper explicitly validates that an FID-reward search yields comparable performance. Thus the headline FID claims are not forced by construction and are checked against external hand-crafted baselines. The transfer of the discovered CIFAR-10 architecture to STL-10 is an independent retraining benchmark. The only partial circularity is the Inception score: because IS is both the reward and the final architecture-selection criterion, the reported IS 8.55 is a selected maximum rather than an unbiased prediction; this is a standard but real overfitting-to-metric issue. The discriminator growth routine is underspecified ('The details of growing D will be explained more in the supplementary') and Section 5 concedes discriminators are not searched, but that is a reproducibility and attribution risk, not circularity: the fixed D schedule is an external design choice, not derived from the results. No load-bearing self-citation, imported uniqueness theorem, or ansatz-smuggling was found. Overall, the central SOTA claim rests on an external metric and survives the circularity check, but the IS headline should be discounted.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new physical entities, particles, or forces. Its contributions are algorithmic: a search space, a controller-reward scheme, a dynamic-resetting heuristic, and a multi-level search variant. The free parameters listed are hand-set hyperparameters that affect the empirical outcome, and the axioms are the domain assumptions about search-space expressiveness, parameter-sharing validity, and the fixed discriminator growth routine.

free parameters (6)
  • dynamic_resetting_variance_threshold = 1e-3
    Hand-chosen threshold on the standard deviation of recent GAN training losses to detect mode collapse and trigger parameter resetting. It affects search efficiency and possibly which architectures are explored.
  • reward_metric = Inception score (IS)
    The controller reward is IS rather than FID. The paper shows FID-based search gives comparable performance, but the choice still shapes which architectures are preferred during search.
  • search_iterations = 90 outer iterations (15 shared-GAN epochs and 30 controller steps each)
    Total search budget is a hand-set hyperparameter. A different budget could yield different architectures and scores.
  • top_K_beams = K (not specified in the main text)
    Beam size for multi-level architecture search appears in Algorithm 1 but the value is not reported. The search result depends on this unspecified hyperparameter.
  • ustage = Not specified in the main text
    Iterations after which the generator and discriminator grow and a new controller is initialized; defined in Algorithm 1 but the exact value is not given, affecting how many cells each controller searches.
  • final_training_iterations = 50,000 generator iterations
    The derived architecture is trained for 50,000 generator iterations; this is a hand-set training budget that influences the final IS and FID scores.
assumptions (5)
  • domain assumption The search space over convolution block type, normalization, upsampling, and skip connections is sufficient to express high-performing unconditional image generators.
    Section 3.1 defines the search space; if it excludes critical operations such as attention or style modules, the discovered architecture cannot reach the true optimum. The paper acknowledges this limitation in Section 5.
  • domain assumption Parameter sharing across child models yields valid approximations of child-model performance for the purpose of guiding the search.
    Section 3.2 builds on ENAS parameter sharing. The proxy-reward correlation in Section 4.3.1 (Spearman 0.779) provides empirical support, but the assumption is not proven in general.
  • domain assumption Inception Score is a suitable reward proxy for final generation quality, including for the FID metric.
    Section 3.2 chooses IS as reward for speed; Section 4.3.2 shows FID-based search is comparable, which supports the choice. Still, optimizing IS does not guarantee optimal FID.
  • domain assumption Growing the discriminator by a fixed pre-defined routine as the generator deepens maintains the balance between generator and discriminator capacities.
    Section 3 states the discriminator is grown according to a routine detailed in the supplementary, which is not present in this preprint. If the routine is unbalanced, the search may fail or produce artifacts not attributable to the NAS method.
  • standard math The hinge adversarial loss with spectral normalization on the discriminator is a stable training configuration for the shared GAN.
    Section 3.3.1 uses this loss following SN-GAN; it is a standard and well-tested configuration in the GAN literature, so this is a reasonable background assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AutoGAN: Neural Architecture Search for Generative Adversarial Networks." pith.science (2026). https://pith.science/paper/2JI2JDEY

@misc{pith2026190803835,
  author       = {Pith},
  title        = {Pith review of: AutoGAN: Neural Architecture Search for Generative Adversarial Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2JI2JDEY}},
  note         = {Machine review of arXiv:1908.03835}
}
read the original abstract

Neural architecture search (NAS) has witnessed prevailing success in image classification and (very recently) segmentation tasks. In this paper, we present the first preliminary study on introducing the NAS algorithm to generative adversarial networks (GANs), dubbed AutoGAN. The marriage of NAS and GANs faces its unique challenges. We define the search space for the generator architectural variations and use an RNN controller to guide the search, with parameter sharing and dynamic-resetting to accelerate the process. Inception score is adopted as the reward, and a multi-level search strategy is introduced to perform NAS in a progressive way. Experiments validate the effectiveness of AutoGAN on the task of unconditional image generation. Specifically, our discovered architectures achieve highly competitive performance compared to current state-of-the-art hand-crafted GANs, e.g., setting new state-of-the-art FID scores of 12.42 on CIFAR-10, and 31.01 on STL-10, respectively. We also conclude with a discussion of the current limitations and future potential of AutoGAN. The code is available at https://github.com/TAMU-VITA/AutoGAN

Figures

Figures reproduced from arXiv: 1908.03835 by the authors.

Figure 1
Figure 1. The running scheme of the RNN controller. At each time step, the controller outputs a hidden vector to be decoded [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The search space of a generator cell in AutoGAN. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The AutoGAN (generator) architecture discov [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (6 more)
Figure 5
Figure 5. Figure 5: The top 2 and top 3 discovered architectures. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 4
Figure 4. Figure 4: The generated CIFAR-10 results of AutoGAN. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 6
Figure 6. Figure 6: The generated STL-10 results of AutoGAN. They [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: The correlation plot of real evaluation and proxy [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 9
Figure 9. Figure 9: Comparing AutoGAN with (blue) and without [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Comparison of AutoGAN with MLAS (blue line) and SLAS (red line) training chemes. 4.3.5 Comparison to random search We implemented the two random search algorithms in [34]: one with weight sharing and the other without weight shar￾ing (early stopping). We re-searched A…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

73 extracted references · 35 canonical work pages

  1. [1]

    MaskConnect - Con- nectivity Learning by Gradient Descent.ECCV, cs.CV , 2018

    Karim Ahmed and Lorenzo Torresani. MaskConnect - Con- nectivity Learning by Gradient Descent.ECCV, cs.CV , 2018

  2. [2]

    Towards principled methods for training generative adversarial networks

    Martin Arjovsky and L ´eon Bottou. Towards principled methods for training generative adversarial networks. arXiv preprint arXiv:1701.04862, 2017

  3. [3]

    Wasserstein generative adversarial networks

    Martin Arjovsky, Soumith Chintala, and L ´eon Bottou. Wasserstein generative adversarial networks. In Interna- tional Conference on Machine Learning , pages 214–223, 2017

  4. [4]

    Designing neural network architectures using rein- forcement learning

    Bowen Baker, Otkrist Gupta, Nikhil Naik, and Ramesh Raskar. Designing neural network architectures using rein- forcement learning. arXiv preprint arXiv:1611.02167, 2016

  5. [5]

    Large scale gan training for high fidelity natural image synthesis

    Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale gan training for high fidelity natural image synthesis. arXiv preprint arXiv:1809.11096, 2018

  6. [6]

    Neural photo editing with introspective adversarial networks

    Andrew Brock, Theodore Lim, James M Ritchie, and Nick Weston. Neural photo editing with introspective adversarial networks. arXiv preprint arXiv:1609.07093, 2016

  7. [7]

    Searching for efficient multi-scale archi- tectures for dense image prediction

    Liang-Chieh Chen, Maxwell Collins, Yukun Zhu, George Papandreou, Barret Zoph, Florian Schroff, Hartwig Adam, and Jon Shlens. Searching for efficient multi-scale archi- tectures for dense image prediction. In Advances in Neural Information Processing Systems, pages 8713–8724, 2018

  8. [8]

    Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolu- tion, and fully connected crfs

    Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin Murphy, and Alan L Yuille. Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolu- tion, and fully connected crfs. IEEE transactions on pattern analysis and machine intelligence, 40(4):834–848, 2018

Show all 73 references
  1. [9]

    Rethinking atrous convolution for seman- tic image segmentation

    Liang-Chieh Chen, George Papandreou, Florian Schroff, and Hartwig Adam. Rethinking atrous convolution for seman- tic image segmentation. arXiv preprint arXiv:1706.05587 , 2017

  2. [10]

    A downsampled variant of imagenet as an alternative to the ci- far datasets

    Patryk Chrabaszcz, Ilya Loshchilov, and Frank Hutter. A downsampled variant of imagenet as an alternative to the ci- far datasets. arXiv preprint arXiv:1707.08819, 2017

  3. [11]

    Generative adversarial nets

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural information processing systems, pages 2672–2680, 2014

  4. [12]

    Class-splitting generative adversarial networks

    Guillermo L Grinblat, Lucas C Uzal, and Pablo M Granitto. Class-splitting generative adversarial networks. arXiv preprint arXiv:1709.07359, 2017

  5. [13]

    Improved training of wasserstein gans

    Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron C Courville. Improved training of wasserstein gans. In Advances in Neural Information Pro- cessing Systems, pages 5767–5777, 2017

  6. [14]

    Prob- gan: Towards probabilistic gan with theoretical guarantees

    Guang-He Lee Yonglong Tian Hao He, Hao Wang. Prob- gan: Towards probabilistic gan with theoretical guarantees. In ICLR, 2019

  7. [15]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

  8. [16]

    Identity mappings in deep residual networks

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual networks. In European conference on computer vision , pages 630–645. Springer, 2016

  9. [17]

    Amc: Automl for model compression and ac- celeration on mobile devices

    Yihui He, Ji Lin, Zhijian Liu, Hanrui Wang, Li-Jia Li, and Song Han. Amc: Automl for model compression and ac- celeration on mobile devices. In European Conference on Computer Vision, pages 815–832. Springer, 2018

  10. [18]

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

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, G ¨unter Klambauer, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a nash equilibrium. arXiv preprint arXiv:1706.08500, 12(1), 2017

  11. [19]

    Mgan: Training generative adversarial nets with multiple generators

    Quan Hoang, Tu Dinh Nguyen, Trung Le, and Dinh Phung. Mgan: Training generative adversarial nets with multiple generators. 2018

  12. [20]

    Long short-term memory

    Sepp Hochreiter and J ¨urgen Schmidhuber. Long short-term memory. Neural computation, 9(8):1735–1780, 1997

  13. [21]

    Batch normalization: Accelerating deep network training by reducing internal co- variate shift

    Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by reducing internal co- variate shift. arXiv preprint arXiv:1502.03167, 2015

  14. [22]

    Image-to-image translation with conditional adversar- ial networks

    Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adversar- ial networks. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages 5967–5976. IEEE, 2017

  15. [23]

    Enlightengan: Deep light enhancement without paired supervision

    Yifan Jiang, Xinyu Gong, Ding Liu, Yu Cheng, Chen Fang, Xiaohui Shen, Jianchao Yang, Pan Zhou, and Zhangyang Wang. Enlightengan: Deep light enhancement without paired supervision. arXiv preprint arXiv:1906.06972, 2019

  16. [24]

    Auto-keras: Ef- ficient neural architecture search with network morphism, 2018

    Haifeng Jin, Qingquan Song, and Xia Hu. Auto-keras: Ef- ficient neural architecture search with network morphism, 2018

  17. [25]

    The relativistic discriminator: a key element missing from standard gan

    Alexia Jolicoeur-Martineau. The relativistic discriminator: a key element missing from standard gan. arXiv preprint arXiv:1807.00734, 2018

  18. [26]

    Progressive growing of gans for improved quality, stability, and variation

    Tero Karras, Timo Aila, Samuli Laine, and Jaakko Lehtinen. Progressive growing of gans for improved quality, stability, and variation. ICLR, 2018

  19. [27]

    A style-based generator architecture for generative adversarial networks

    Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. arXiv preprint arXiv:1812.04948, 2018

  20. [28]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 , 2014

  21. [29]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. Technical report, Cite- seer, 2009

  22. [30]

    Imagenet classification with deep convolutional neural net- works

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural net- works. In Advances in neural information processing sys- tems, pages 1097–1105, 2012

  23. [31]

    Deblurgan: Blind motion deblurring using conditional adversarial networks

    Orest Kupyn, V olodymyr Budzan, Mykola Mykhailych, Dmytro Mishkin, and Ji ˇr´ı Matas. Deblurgan: Blind motion deblurring using conditional adversarial networks. In Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 8183–8192, 2018

  24. [32]

    The gan landscape: Losses, archi- tectures, regularization, and normalization

    Karol Kurach, Mario Lucic, Xiaohua Zhai, Marcin Michal- ski, and Sylvain Gelly. The gan landscape: Losses, archi- tectures, regularization, and normalization. arXiv preprint arXiv:1807.04720, 2018

  25. [33]

    Photo- realistic single image super-resolution using a generative ad- versarial network

    Christian Ledig, Lucas Theis, Ferenc Husz´ar, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, et al. Photo- realistic single image super-resolution using a generative ad- versarial network. In 2017 IEEE Conference ...

  26. [34]

    Random search and repro- ducibility for neural architecture search

    Liam Li and Ameet Talwalkar. Random search and repro- ducibility for neural architecture search. arXiv, 2019

  27. [35]

    Feature pyramid networks for object detection

    Tsung-Yi Lin, Piotr Doll ´ar, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyramid networks for object detection. In CVPR, volume 1, page 4, 2017

  28. [36]

    Auto-deeplab: Hierarchical neural architecture search for semantic image segmentation

    Chenxi Liu, Liang-Chieh Chen, Florian Schroff, Hartwig Adam, Wei Hua, Alan Yuille, and Li Fei-Fei. Auto-deeplab: Hierarchical neural architecture search for semantic image segmentation. arXiv preprint arXiv:1901.02985, 2019

  29. [37]

    Progressive Neural Architecture Search

    Chenxi Liu, Barret Zoph, Maxim Neumann, Jonathon Shlens, Wei Hua, Li-Jia Li, Li Fei-Fei, Alan Yuille, Jonathan Huang, and Kevin Murphy. Progressive Neural Architecture Search. pages 19–34, 2018

  30. [38]

    Darts: Differentiable architecture search

    H Liu, K Simonyan, Y Yang arXiv preprint arXiv 1806.09055, and 2018. Darts: Differentiable architecture search. arxiv.org

  31. [39]

    Are gans created equal? a large-scale study

    Mario Lucic, Karol Kurach, Marcin Michalski, Sylvain Gelly, and Olivier Bousquet. Are gans created equal? a large-scale study. In Advances in neural information pro- cessing systems, pages 700–709, 2018

  32. [40]

    Multi-task sequence to sequence learning

    Minh-Thang Luong, Quoc V Le, Ilya Sutskever, Oriol Vinyals, and Lukasz Kaiser. Multi-task sequence to sequence learning. arXiv preprint arXiv:1511.06114, 2015

  33. [41]

    Conditional generative adversarial nets

    Mehdi Mirza and Simon Osindero. Conditional generative adversarial nets. arXiv preprint arXiv:1411.1784, 2014

  34. [42]

    Spectral normalization for generative ad- versarial networks

    Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for generative ad- versarial networks. arXiv preprint arXiv:1802.05957, 2018

  35. [43]

    Stacked hour- glass networks for human pose estimation

    Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked hour- glass networks for human pose estimation. InEuropean Con- ference on Computer Vision, pages 483–499. Springer, 2016

  36. [44]

    Dual dis- criminator generative adversarial nets

    Tu Nguyen, Trung Le, Hung Vu, and Dinh Phung. Dual dis- criminator generative adversarial nets. InAdvances in Neural Information Processing Systems, pages 2670–2680, 2017

  37. [45]

    De- convolution and checkerboard artifacts

    Augustus Odena, Vincent Dumoulin, and Chris Olah. De- convolution and checkerboard artifacts. Distill, 1(10):e3, 2016

  38. [46]

    Efficient Neural Architecture Search via Parameter Sharing

    H Pham, M Y Guan, B Zoph, Q V Le, J Dean arXiv preprint arXiv, and 2018. Efficient Neural Architecture Search via Parameter Sharing. arxiv.org

  39. [47]

    Un- supervised representation learning with deep convolu- tional generative adversarial networks

    Alec Radford, Luke Metz, and Soumith Chintala. Un- supervised representation learning with deep convolu- tional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015

  40. [48]

    Generative adversarial text to image synthesis

    S Reed, Z Akata, X Yan, L Logeswaran arXiv preprint arXiv, and 2016. Generative adversarial text to image synthesis. jmlr.org, 2016

  41. [49]

    U- net: Convolutional networks for biomedical image segmen- tation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U- net: Convolutional networks for biomedical image segmen- tation. In International Conference on Medical image com- puting and computer-assisted intervention , pages 234–241. Springer, 2015

  42. [50]

    Improved techniques for training gans

    Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training gans. In Advances in Neural Information Pro- cessing Systems, pages 2234–2242, 2016

  43. [51]

    Going deeper with convolutions

    Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1–9, 2015

  44. [52]

    Rethinking the inception archi- tecture for computer vision

    Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the inception archi- tecture for computer vision. In Proceedings of the IEEE con- ference on computer vision and pattern recognition , pages 2818–2826, 2016

  45. [53]

    Deep and hierarchical implicit models

    Dustin Tran, Rajesh Ranganath, and David M Blei. Deep and hierarchical implicit models. CoRR, abs/1702.08896, 2017

  46. [54]

    Dist-gan: An improved gan using distance constraints

    Ngoc-Trung Tran, Tuan-Anh Bui, and Ngai-Man Cheung. Dist-gan: An improved gan using distance constraints. In Proceedings of the European Conference on Computer Vi- sion (ECCV), pages 370–385, 2018

  47. [55]

    In- stance normalization: The missing ingredient for fast styliza- tion

    Dmitry Ulyanov, Andrea Vedaldi, and Victor Lempitsky. In- stance normalization: The missing ingredient for fast styliza- tion. arXiv preprint arXiv:1607.08022, 2016

  48. [56]

    Improv- ing mmd-gan training with repulsive loss function

    Wei Wang, Yuan Sun, and Saman Halgamuge. Improv- ing mmd-gan training with repulsive loss function. arXiv preprint arXiv:1812.09916, 2018

  49. [57]

    Studying very low resolution recogni- tion using deep networks

    Zhangyang Wang, Shiyu Chang, Yingzhen Yang, Ding Liu, and Thomas S Huang. Studying very low resolution recogni- tion using deep networks. In Proceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition, pages 4792–4800, 2016

  50. [58]

    Improving gen- erative adversarial networks with denoising feature match- ing.(2017)

    David Warde-Farley and Yoshua Bengio. Improving gen- erative adversarial networks with denoising feature match- ing.(2017). In ICLR, 2017

  51. [59]

    Simple statistical gradient-following al- gorithms for connectionist reinforcement learning

    Ronald J Williams. Simple statistical gradient-following al- gorithms for connectionist reinforcement learning. Machine learning, 8(3-4):229–256, 1992

  52. [60]

    Genetic cnn

    Lingxi Xie and Alan Yuille. Genetic cnn. In 2017 IEEE International Conference on Computer Vision (ICCV), pages 1388–1397. IEEE, 2017

  53. [61]

    Attngan: Fine- grained text to image generation with attentional generative adversarial networks

    Tao Xu, Pengchuan Zhang, Qiuyuan Huang, Han Zhang, Zhe Gan, Xiaolei Huang, and Xiaodong He. Attngan: Fine- grained text to image generation with attentional generative adversarial networks. arXiv preprint, 2017

  54. [62]

    Lr-gan: Layered recursive generative adver- sarial networks for image generation

    Jianwei Yang, Anitha Kannan, Dhruv Batra, and Devi Parikh. Lr-gan: Layered recursive generative adver- sarial networks for image generation. arXiv preprint arXiv:1703.01560, 2017

  55. [63]

    Controllable artistic text style transfer via shape-matching gan

    Shuai Yang, Zhangyang Wang, Zhaowen Wang, Ning Xu, Jiaying Liu, and Zongming Guo. Controllable artistic text style transfer via shape-matching gan. arXiv preprint arXiv:1905.01354, 2019

  56. [64]

    Self-attention generative adversarial networks

    Han Zhang, Ian Goodfellow, Dimitris Metaxas, and Augus- tus Odena. Self-attention generative adversarial networks. arXiv preprint arXiv:1805.08318, 2018

  57. [65]

    Stack- gan++: Realistic image synthesis with stacked generative ad- versarial networks

    Han Zhang, Tao Xu, Hongsheng Li, Shaoting Zhang, Xi- aogang Wang, Xiaolei Huang, and Dimitris Metaxas. Stack- gan++: Realistic image synthesis with stacked generative ad- versarial networks. arXiv preprint arXiv:1710.10916, 2017

  58. [66]

    Stack- gan: Text to photo-realistic image synthesis with stacked generative adversarial networks

    Han Zhang, Tao Xu, Hongsheng Li, Shaoting Zhang, Xiao- gang Wang, Xiaolei Huang, and Dimitris N Metaxas. Stack- gan: Text to photo-realistic image synthesis with stacked generative adversarial networks. In Proceedings of the IEEE International Conference on Computer Vision, pa...

  59. [67]

    Dada: Deep adversarial data augmentation for ex- tremely low data regime classification

    Xiaofeng Zhang, Zhangyang Wang, Dong Liu, and Qing Ling. Dada: Deep adversarial data augmentation for ex- tremely low data regime classification. In ICASSP 2019- 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages 2807–2811. IEEE, 2019

  60. [68]

    Energy- based generative adversarial network

    Junbo Zhao, Michael Mathieu, and Yann LeCun. Energy- based generative adversarial network. arXiv preprint arXiv:1609.03126, 2016

  61. [69]

    Practical block-wise neural network architecture gener- ation

    Zhao Zhong, Junjie Yan, Wei Wu, Jing Shao, and Cheng-Lin Liu. Practical block-wise neural network architecture gener- ation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 2423–2432, 2018

  62. [70]

    Unpaired Image-to-Image Translation using Cycle- Consistent Adversarial Networks

    Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired Image-to-Image Translation using Cycle- Consistent Adversarial Networks. arXiv.org, Mar. 2017

  63. [71]

    Neural architecture search with reinforcement learning

    Barret Zoph and Quoc V Le. Neural architecture search with reinforcement learning. arXiv preprint arXiv:1611.01578 , 2016

  64. [72]

    Learning Transferable Architectures for Scalable Image Recognition

    Barret Zoph, Vijay Vasudevan, Jonathon Shlens, and Quoc V Le. Learning Transferable Architectures for Scalable Image Recognition. CVPR, 2018

  65. [73]

    Transfer learning for low-resource neural machine transla- tion

    Barret Zoph, Deniz Yuret, Jonathan May, and Kevin Knight. Transfer learning for low-resource neural machine transla- tion. arXiv preprint arXiv:1604.02201, 2016

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.