Training Wasserstein GANs without gradient penalties

Kwon, Dohyun; Kim, Yeoneung; Montúfar, Guido; Yang, Insoon

doi:10.48550/arxiv.2110.14150

Cited by 2 publications

(2 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Additionally, we believe that the procedure outlined in this paper could be applied to augment analytical approximations to 𝑁-body simulations (like L-PICOLA, Howlett et al 2015, or FastPM, Feng et al 2016, as well as semi-analytical models of galaxies, which, in the same vein as lognormal random fields, provide a fast approximation to hydrodynamical simulations by modelling complicated baryonic processes (White & Frenk 1991;Kauffmann et al 1993;Cole et al 1994;Somerville & Primack 1999;Lacey 2001). We further plan to explore the possibility to employ the dataset described in this work to reduce the variance in the statistics of large-scale structure observables using a small number of expensive simulations Ding et al 2022), as well as to replace our WGAN-GP model with either a possibly more stable GAN version (Kwon et al 2021), or with a more compact model, like the one proposed in the context of Lagrangian deep learning (LDL, Dai & Seljak 2021), using graph neural networks (GNNs, see e.g. Zhou et al 2018 for a review) or through normalising flows (e.g.…”

Section: Discussionmentioning

confidence: 99%

Fast and realistic large-scale structure from machine-learning-augmented random field simulations

Piras¹,

Joachimi²,

Villaescusa-Navarro³

2022

Preprint

View full text Add to dashboard Cite

Producing thousands of simulations of the dark matter distribution in the Universe with increasing precision is a challenging but critical task to facilitate the exploitation of current and forthcoming cosmological surveys. Many inexpensive substitutes to full 𝑁-body simulations have been proposed, even though they often fail to reproduce the statistics of the smaller, non-linear scales. Among these alternatives, a common approximation is represented by the lognormal distribution, which comes with its own limitations as well, while being extremely fast to compute even for high-resolution density fields. In this work, we train a machine learning model to transform projected lognormal dark matter density fields to more realistic dark matter maps, as obtained from full 𝑁-body simulations. We detail the procedure that we follow to generate highly correlated pairs of lognormal and simulated maps, which we use as our training data, exploiting the information of the Fourier phases. We demonstrate the performance of our model comparing various statistical tests with different field resolutions, redshifts and cosmological parameters, proving its robustness and explaining its current limitations. The augmented lognormal random fields reproduce the power spectrum up to wavenumbers of 1 ℎ Mpc −1 , the bispectrum and the peak counts within 10%, and always within the error bars, of the fiducial target simulations. Finally, we describe how we plan to integrate our proposed model with existing tools to yield more accurate spherical random fields for weak lensing analysis, going beyond the lognormal approximation.

show abstract

Section: Discussionmentioning

confidence: 99%

Fast and realistic large-scale structure from machine-learning-augmented random field simulations

Piras¹,

Joachimi²,

Villaescusa-Navarro³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…The excellent auxiliary classifier GAN model [28] that appeared recently improved and optimized the unstable training of adversarial networks. Later, the WGAN (Wasserstein GAN) was proposed, which replaced Jensen-Shannon divergence and Kullback-Leibler divergence [19,29] with Earth Mover's Distance [30] and completely solved the instability problem of GAN training. Utilizing the gradient penalty (GP) method in WGAN-GP [31] can improve the training effect and speed up the convergence speed.…”

Section: Introductionmentioning

confidence: 99%

Improved CycleGAN for super-resolution of engineering drawings

Song

Yao²,

Tian³

et al. 2023

Meas. Sci. Technol.

View full text Add to dashboard Cite

In intelligent manufacturing, the quality of machine translation engineering drawings will directly affect its manufacturing accuracy. Currently, most of the engineering drawing translation work is done manually, which greatly reduces production efficiency. This paper proposes an automatic translation method for welded structural engineering drawings based on Cyclic Generative Adversarial Networks (Cyclegan). The Cyclegan network model of unpaired transfer learning is used to learn the feature mapping of real welding engineering drawings to realize the automatic translation of engineering drawings. U-Net and Patchgan are the main network for the generator and discriminator, respectively. Based on removing the identity mapping function, a high-dimensional sparse network is proposed to replace the traditional dense network for the Cyclegan generator to improve noise robustness. Increase the residual block hidden layer to increase the resolution of the generated graph. The improved and fine-tuned network models are experimentally validated, computing the gap between real and generated data. It meets the welding engineering precision standard and solves the main problem of low drawing recognition efficiency in the welding manufacturing process. The results show. After training with our model, the PSNR, SSIM, and MSE of welding engineering drawings reach about 44.89%, 99.58%, and 2.11, respectively, which are superior to traditional networks in both training speed and accuracy.

show abstract

Training Wasserstein GANs without gradient penalties

Cited by 2 publications

References 4 publications

Fast and realistic large-scale structure from machine-learning-augmented random field simulations

Fast and realistic large-scale structure from machine-learning-augmented random field simulations

Improved CycleGAN for super-resolution of engineering drawings

Contact Info

Product

Resources

About