Loss-Sensitive Generative Adversarial Networks on Lipschitz Densities

By Guo-Jun Qi. 2017

365 citations recorded by the source corpus.

In this paper, we present the Lipschitz regularization theory and algorithms for a novel Loss-Sensitive Generative Adversarial Network (LS-GAN). Specifically, it trains a loss function to distinguish between real and fake samples by designated margins, while learning a generator alternately to produce realistic samples by minimizing their losses. The LS-GAN further regularizes its loss function with a Lipschitz regularity condition on the density of real data, yielding a regularized model that can better generalize to produce new data from a reasonable number of training examples than the classic GAN. We will further present a Generalized LS-GAN (GLS-GAN) and show it contains a large family of regularized GAN models, including both LS-GAN and Wasserstein GAN, as its special cases. Compared with the other GAN models, we will conduct experiments to show both LS-GAN and GLS-GAN exhibit competitive ability in generating new images in terms of the Minimum Reconstruction Error (MRE) assessed on a separate test set. We further extend the LS-GAN to a conditional form for supervised and semi-supervised learning problems, and demonstrate its outstanding performance on image classification tasks.

Subfields: Computer Vision and Pattern Recognition, Generative Models

How researchers see this paper

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

In Computer Vision and Pattern Recognition

In Generative Models

Citations in the corpus

Spectral Normalization for Generative Adversarial Networks (2018, background): “A number of works (Ue-hara et al., 2016; Qi, 2017; Gulrajani et al., 2017) advocate the importance of Lipschitz continuity in assuring the boundedness of statistics.”
Generative Adversarial Networks (2021, neutral): “In previous works that stabilized GAN training ([21, 4, 56]) it was emphasized that D should be a K-Lipshitz continuous function, forcing it not to change rapidly.”
Least Squares Generative Adversarial Networks (2016, background): “Many works have been proposed to address this problem by analyzing the objective functions of GANs [2, 3, 19, 23, 24].”
Geometric GAN (2017, methodology): “In this paper, we only consider Lipschitz density constraints in [18], so we follow to use weight decay on generators g θ ( x ) and feature space mapping φ ζ ( x ) in discriminators.”
Recent Progress on Generative Adversarial Networks (GANs): A Survey (2019, methodology): “Both [25] and [27] used the Lipschitz regularity to address the problem of mode collapse and vanishing gradient.”
AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transformations Rather Than Data (2019, background): “Recent progress has shown the promising generalization ability of regularized GANs in generating unseen data based on the Lipschitz assumption on the real data distribution [23, 2], and this shows great potential of GANs in providing expressive representation of images [6, 8, 9].”
Robust Imitation of Diverse Behaviors (2017, background): “Recent work has focused on alleviating mode collapse in image modeling [2, 4, 19, 25, 41, 11], but so far these have not been demonstrated in the control domain.”
An empirical study on evaluation metrics of generative adversarial networks (2018, background): “However, despite the availability of a plethora of GAN models (Arjovsky et al., 2017; Qi, 2017; Zhao et al., 2016), their evaluation is still predominantly qualitative, very often resorting to manual inspection of the visual ﬁdelity of generated images.”
How to Train Your DRAGAN (2017, background): “Guo-Jun Qi [20] introduced LS-GAN with the idea of maintaining a margin between losses assigned to real and fake samples.”
A Survey on Generative Adversarial Networks: Variants, Applications, and Training (2020, background): “) [225], Kernel Inception Distance (KID) [226], Fréchet Inception Distance (FID) [198], Multi-Scale Structural Similarity (MS-SSIM) [227], Classifier Two-Sample Tests (C2ST) [228], Wasserstein Critic [27], and Maximum Mean Discrepancy (MMD) [229], there are no standard consensus parameters for fair ”
On Convergence and Stability of GANs (2018, background): “This partly explains the success of WGAN and gradient penalties in the recent literature (Gulrajani et al., 2017; Qi, 2017), and why they improve the training stability of GANs, despite being motivated by reasoning based on unrealistic assumptions.”
How Generative Adversarial Networks and Their Variants Work (2017, methodology): “Loss sensitive GAN (LS-GAN) [98] also uses a Lipschitz constraint but with a diﬀerent method.”
Towards the Automatic Anime Characters Creation with Generative Adversarial Networks (2017, background): “By learning a loss function to separate generated samples from their real examples, LS-GAN[21] focuses on improving poor generation result and thus avoids mode collapse.”
Pose Transferrable Person Re-identification (2018, background): “On one hand, many works explore to improve model structure and optimization form [26, 1, 25, 28].”
How Generative Adversarial Networks and Its Variants Work : An Overview of GAN (2017, methodology): “Moreover, LS-GAN assumes the density of real samples pdata(x) is Lipschitz continuous, so that nearby data does not abruptly change.”
Deep Learning in Mobile and Wireless Networking: A Survey (2018, methodology): “There exist several solutions for mitigating this problem, e.g., Wasserstein Generative Adversarial Network (WGAN) [79], Loss-Sensitive Generative Adversarial Network (LS-GAN) [168] and BigGAN [169], but research on the theory of GANs remains shallow.”