Semi-Supervised Learning with GANs: Revisiting Manifold Regularization

Lecouat, Bruno; Foo, Chuan-Sheng; Zenati, Houssam; Chandrasekhar, Vijay

doi:10.48550/arxiv.1805.08957

Cited by 10 publications

(19 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Method CIFAR-10 4,000 labels VAT [18] 13.15 VAT + SNTG [15] 12.49 Π model [11] 16.55 Mean Teacher [24] 17.74 CCLP [8] 18.57 ALI [3] 17.99 Improved GAN [21] 18.63 Tripple GAN [14] 16.99 Bad GAN [2] 14.41 LGAN [20] 14.23 Improved GAN + JacobRegu + tangent [10] 16.20 Improved GAN + ManiReg [12] 14.45 TNAR [28] 12.06 Pani VAT (input) 12.20 Pani VAT (+hidden)…”

Section: Pani Vatmentioning

confidence: 99%

Patch-level Neighborhood Interpolation: A General and Effective Graph-based Regularization Strategy

Sun¹,

Yu²,

Lin³

et al. 2019

Preprint

View full text Add to dashboard Cite

Regularization plays a crucial role in machine learning models, especially for deep neural networks. The existing regularization techniques mainly reply on the i.i.d. assumption and only employ the information of the current sample, without the leverage of neighboring information between samples. In this work, we propose a general regularizer called Patch-level Neighborhood Interpolation (Pani) that fully exploits the relationship between samples. Furthermore, by explicitly constructing a patch-level graph in the different network layers and interpolating the neighborhood features to refine the representation of the current sample, our Patch-level Neighborhood Interpolation can then be applied to enhance two popular regularization strategies, namely Virtual Adversarial Training (VAT) and MixUp, yielding their neighborhood versions. The first derived Pani VAT presents a novel way to construct nonlocal adversarial smoothness by incorporating patch-level interpolated perturbations. In addition, the Pani MixUp method extends the original MixUp regularization to the patch level and then can be developed to MixMatch, achieving the stateof-the-art performance. Finally, extensive experiments are conducted to verify the effectiveness of the Patch-level Neighborhood Interpolation in both supervised and semi-supervised settings.

show abstract

Section: Pani Vatmentioning

confidence: 99%

Patch-level Neighborhood Interpolation: A General and Effective Graph-based Regularization Strategy

Sun¹,

Yu²,

Lin³

et al. 2019

Preprint

View full text Add to dashboard Cite

show abstract

“…Several optimization heuristics and architectures have also been proposed to address challenges such as mode collapse [25,16,22,5]. Methods for regularizing the discriminator for better stability were devised in [23,15,17,11]. The authors in [23] presented a stabilizing regularizer that is based on a gradient norm, where the gradient is calculated with respect to the data samples.…”

Section: Related Workmentioning

confidence: 99%

“…On the other hand, the authors of [15,17] designed regularizers based on the norm of a gradient calculated with respect to the parameters. The authors in [11] applied a Jacobian regularizer to the discriminator of a featurematching GAN to improve the performance of GAN-based semi-supervised learning. In contrast to regularizing the discriminator, this paper proposes to regularize the generator for improving GAN training.…”

Section: Related Workmentioning

confidence: 99%

MR-GAN: Manifold Regularized Generative Adversarial Networks

Li,

Kailkhura,

Anirudh

et al. 2018

Preprint

View full text Add to dashboard Cite

Despite the growing interest in generative adversarial networks (GANs), training GANs remains a challenging problem, both from a theoretical and a practical standpoint. To address this challenge, in this paper, we propose a novel way to exploit the unique geometry of the real data, especially the manifold information. More specifically, we design a method to regularize GAN training by adding an additional regularization term referred to as manifold regularizer. The manifold regularizer forces the generator to respect the unique geometry of the real data manifold and generate high quality data. Furthermore, we theoretically prove that the addition of this regularization term in any class of GANs including DCGAN and Wasserstein GAN leads to improved performance in terms of generalization, existence of equilibrium, and stability. Preliminary experiments show that the proposed manifold regularization helps in avoiding mode collapse and leads to stable training.

show abstract

“…These recent advances of generative modeling have shown huge potential to many computer vision tasks, e.g. image inpainting [61,40,55,54], semi-supervised learning [13,46,27,4,49,4,2], data manipulation [9,60], high-resolution image generation [29,23,48,39], transfer learning [7,51,20,47], image-to-image translation [61,56,30,11,62,21], and textto-image [43,44,57,53,58], to name a few.…”

Section: Introductionmentioning

confidence: 99%

Adversarial Code Learning for Image Generation

Yuan,

Wu,

Ding

et al. 2020

Preprint

View full text Add to dashboard Cite

We introduce the "adversarial code learning" (ACL) module that improves overall image generation performance to several types of deep models. Instead of performing a posterior distribution modeling in the pixel spaces of generators, ACLs aim to jointly learn a latent code with another image encoder/inference net, with a prior noise as its input. We conduct the learning in an adversarial learning process, which bears a close resemblance to the original GAN but again shifts the learning from image spaces to prior and latent code spaces. ACL is a portable module that brings up much more flexibility and possibilities in generative model designs. First, it allows flexibility to convert non-generative models like Autoencoders and standard classification models to decent generative models. Second, it enhances existing GANs' performance by generating meaningful codes and images from any part of the priori. We have incorporated our ACL module with the aforementioned frameworks and have performed experiments on synthetic, MNIST, CIFAR-10, and CelebA datasets. Our models have achieved significant improvements which demonstrated the generality for image generation tasks 1 .

show abstract

Semi-Supervised Learning with GANs: Revisiting Manifold Regularization

Cited by 10 publications

References 0 publications

Patch-level Neighborhood Interpolation: A General and Effective Graph-based Regularization Strategy

Patch-level Neighborhood Interpolation: A General and Effective Graph-based Regularization Strategy

MR-GAN: Manifold Regularized Generative Adversarial Networks

Adversarial Code Learning for Image Generation

Contact Info

Product

Resources

About