On Convergence and Stability of GANs

By Naveen Kodali, James Hays, Jacob D. Abernethy, Zsolt Kira. 2018

666 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Generative Models

Citations in the corpus

Are GANs Created Equal? A Large-Scale Study (2017, methodology): “Gradient norm penalty can also be added to both MM GAN and NS GAN and evaluated around the data manifold (DRAGAN [15] in Table 1 based on NS GAN).”
Large Scale GAN Training for High Fidelity Natural Image Synthesis (2018, background): “Another line is focused on constraining D through gradient penalties (Gulrajani et al., 2017; Kodali et al., 2017; Mescheder et al., 2018) or normalization (Miyato et al.”
Diagnosing and Enhancing VAE Models (2019, background): “7 Additionally, we present results from (Lucic et al., 2018) involving numerous competing GAN models, including MM GAN (Goodfellow et al., 2014), WGAN (Arjovsky et al., 2017), WGAN-GP (Gulrajani et al., 2017), NS GAN (Fedus et al., 2017), DRAGAN (Ko-dali et al., 2017), LS GAN (Mao et al., 2017) and ”
Sorting out Lipschitz function approximation (2018, neutral): “, 2018) and stabilizing GAN training (Kodali et al., 2017).”
The GAN Landscape: Losses, Architectures, Regularization, and Normalization (2018, methodology): “…architectures described in Table 4a. Finally, we consider batch normalization (BN) (Ioffe and Szegedy, 2015), layer normalization (LN) (Ba et al., 2016), spectral normalization (SN), gradient penalty (GP) (Gulrajani et al., 2017), dragan penalty (DR) (Kodali et al., 2017), or L 2 regularization.”
GANSynth: Adversarial Neural Audio Synthesis (2019, background): “…Generative Adversarial Networks (GANs) (Goodfellow et al., 2014) have seen great recent success at generating high resolution images (Radford et al., 2016; Arjovsky et al., 2017; Gulrajani et al., 2017; Berthelot et al., 2017; Kodali et al., 2017; Karras et al., 2018a; Miyato et al., 2018).”
Adversarial score matching and improved sampling for image generation (2020, background): “These ﬁndings imply that score matching performs better than most GANs and on-par with state-of-the-art GANs.”
Consistency Regularization for Generative Adversarial Networks (2019, methodology): “First we conduct a large scale study to compare consistency regularization to existing GAN regularization techniques (Kodali et al., 2017; Gul-rajani et al., 2017; Roth et al., 2017) for several GAN architectures, loss functions and other hyper-parameter settings.”
Generative Adversarial Network in Medical Imaging: A Review (2018, methodology): “It exhibited good performance and could potentially replace other regularization techniques, such as gradient penalty regularization [47], Dragan regularization [72].”
Adversarial Transfer Learning (2018, methodology): “[64] used a similar gradient penalty in their deep regret analytic GAN (DRAGAN) training method, which helped to avoid mode collapse.”
Feature Quantization Improves GAN Training (2020, methodology): “Various Regularization methods have been proposed, including changing the objective functions to encourage convergence (Arjovsky et al., 2017; Mao et al., 2017; Mescheder et al., 2018; Kodali et al., 2017), and constraining discriminator through gradient penalties (Gulrajani et al.”
Generative artificial intelligence (2023, positive): “They also feature improved stability over other DGMs during training to be less prone to issues like mode collapse (Kodali et al., 2017; Rombach et al., 2022), making them well-suited for high-quality and detailed outputs, such as high-resolution image synthesis (Ho et al., 2020) to networked…”
On the Effectiveness of Least Squares Generative Adversarial Networks (2017, methodology): “Gradient penalty has been proven to be effective in improving the stability of GAN training [18], [19].”
RELATE: Physically Plausible Multi-Object Scene Synthesis Using Structured Latent Spaces (2020, background): “such as DCGAN[30] and DRAGAN[20], and object-centric generative baselines GENESIS [7] and OCF [1], quoting results from the original papers whenever possible.”
Learning Implicit Generative Models with the Method of Learned Moments (2018, methodology): “The results reported for DRAGAN, GAN-GP, and WGAN-GP were the best obtained in a hyperparameter sweep over discriminator learning rates in 0.0001, 0.0002, 0.0003 and generator learning rates in the same interval.”
Global Texture Enhancement for Fake Face Detection in the Wild (2020, methodology): “Early works [8, 29, 15, 1, 2] can generate high quality low resolution images but suffer from mode collapse issues for generating high resolution images.”