Filters








7,429 Hits in 2.6 sec

Projected GANs Converge Faster [article]

Axel Sauer, Kashyap Chitta, Jens Müller, Andreas Geiger
2021 arXiv   pre-print
Our Projected GAN improves image quality, sample efficiency, and convergence speed.  ...  Importantly, Projected GANs match the previously lowest FIDs up to 40 times faster, cutting the wall-clock time from 5 days to less than 3 hours given the same computational resources.  ...  We implement these baselines and our Projected GANs within the codebase provided by the authors of StyleGAN2-ADA [32] .  ... 
arXiv:2111.01007v1 fatcat:zelvwjh46bcftkywsuota7im4u

Supplementary Information: Adaptive Partial Scanning Transmission Electron Microscopy with Reinforcement Learning [article]

Jeffrey Ede
2020 Zenodo  
S2d show that MSEs result in faster convergence than maximum region Figure S2 .  ...  Adding a projecting layer increases the initial rate of convergence; however, it also increases final losses.  ... 
doi:10.5281/zenodo.4304453 fatcat:22kdnxnpxzaxbbpdopszjf2f44

Learning Constrained Distributions of Robot Configurations with Generative Adversarial Network [article]

Teguh Santoso Lembono, Emmanuel Pignat, Julius Jankowski, Sylvain Calinon
2021 arXiv   pre-print
Then, we use it to generate samples in sampling-based constrained motion planning algorithms to reduce the necessary projection steps, speeding up the computation.  ...  We can see that using GAN speeds up both projection and IK computation significantly, around 2-5 times faster than uniform sampling, even when considering only the successful results.  ...  GAN samples only require around 2-4 optimization steps to achieve convergence.  ... 
arXiv:2011.05717v2 fatcat:c5bz35z2qvc5zhbhe6yo5ofhfy

GAN-based Projector for Faster Recovery with Convergence Guarantees in Linear Inverse Problems [article]

Ankit Raj, Yuqi Li, Yoram Bresler
2019 arXiv   pre-print
Here, we propose a new method of deploying a GAN-based prior to solve linear inverse problems using projected gradient descent (PGD).  ...  Because the learning of the GAN and projector is decoupled from the measurement operator, our GAN-based projector and recovery algorithm are applicable without retraining to all linear inverse problems  ...  Conclusion In this work, we propose a GAN based projection network for faster recovery in linear inverse problems.  ... 
arXiv:1902.09698v2 fatcat:m7oqxhwp2rchhgzv3df62wx22i

Language Modeling with Generative Adversarial Networks [article]

Mehrad Moradshahi, Utkarsh Contractor
2018 arXiv   pre-print
Furthermore, we present the results of some experiments that indicate better training and convergence of Wasserstein GANs (WGANs) when a weaker regularization term is enforcing the Lipschitz constraint  ...  GANs were originally designed to output differentiable values, so discrete language generation is challenging for them which causes high levels of instability in training GANs.  ...  It trains faster than GAN and produces more coherent results. Even though it still produce illegible phrases occasionally, it learns much faster not to repeat the same mistakes again.  ... 
arXiv:1804.02617v1 fatcat:uboxaxkxujbplgtgw5u7gnynsy

GAN-Based Projector for Faster Recovery With Convergence Guarantees in Linear Inverse Problems

Ankit Raj, Yuqi Li, Yoram Bresler
2019 2019 IEEE/CVF International Conference on Computer Vision (ICCV)  
Here, we propose a new method of deploying a GAN-based prior to solve linear inverse problems using projected gradient descent (PGD).  ...  Because the learning of the GAN and projector is decoupled from the measurement operator, our GAN-based projector and recovery algorithm are applicable without retraining to all linear inverse problems  ...  Comparison of Run-time for Recovery Analysis: Error in Projector Conclusion In this work, we propose a GAN based projection network for faster recovery in linear inverse problems.  ... 
doi:10.1109/iccv.2019.00570 dblp:conf/iccv/RajLB19 fatcat:bop37m6m4fdrrjeh4anobb553q

Training Generative Adversarial Networks with Adaptive Composite Gradient [article]

Huiqing Qi, Fang Li, Shengli Tan, Xiangyun Zhang
2021 arXiv   pre-print
Theory and toy-function experiments suggest that our approach can alleviate the cyclic behaviors and converge faster than recently proposed algorithms.  ...  We conducted two mixture of Gaussians experiments by integrating ACG to existing algorithms with Linear GANs. Results show ACG is competitive with the previous algorithms.  ...  However, the other methods all converge to the Nash Equilibrium. We proposed the ACG method converges faster than other convergence methods.  ... 
arXiv:2111.05508v1 fatcat:x4ymv2zt6nc6xdxmgyne6e75p4

Megapixel Size Image Creation using Generative Adversarial Networks [article]

Marco Marchesi
2017 arXiv   pre-print
In image generation several projects showed how GANs are able to generate photorealistic images but the results so far did not look adequate for the quality standard of visual media production industry  ...  Since its appearance, Generative Adversarial Networks (GANs) have received a lot of interest in the AI community.  ...  Acknowledgement This research was part of a commercial project funded by MHPC.  ... 
arXiv:1706.00082v1 fatcat:wahjgbltbzdtfl77pvyg35zm24

Fast learning from label proportions with small bags [article]

Denis Baručić
2021 arXiv   pre-print
In comparison to existing deep LLP methods, our approach converges faster to a comparable or better solution. Several experiments were performed on two different datasets.  ...  Acknowledgements The authors acknowledge the support of the OP VVV funded project "CZ.02.1.01/0.0/0.0/16_019/0000765 Research Center for Informatics" and the Grant Agency of the Czech Technical University  ...  Faster convergence is especially desirable when big datasets consisting of highdimensional instances are processed.  ... 
arXiv:2110.03426v2 fatcat:3pym4sw5kvacvdun2rfa76psze

Inverse mapping of face GANs [article]

Nicky Bayat, Vahid Reza Khazaie, Yalda Mohsenzadeh
2020 arXiv   pre-print
Generative adversarial networks (GANs) synthesize realistic images from a random latent vector.  ...  In addition, our proposed method projects generated faces to their latent-space with high fidelity and speed. At last, we demonstrate the performance of our approach on both real and generated faces.  ...  Methods focusing on projecting images into the corresponding GAN latent vectors can be categorized into four major groups.  ... 
arXiv:2009.05671v1 fatcat:h744nq724zblflol6pk4jeatme

A study on the use of Boundary Equilibrium GAN for Approximate Frontalization of Unconstrained Faces to aid in Surveillance [article]

Wazeer Zulfikar, Sebastin Santy, Sahith Dambekodi, Tirtharaj Dash
2018 arXiv   pre-print
This approach could produce a promising output along with a faster and more stable training for the model.  ...  The BEGAN model additionally has a balanced generator-discriminator combination, which prevents mode collapse along with a global convergence measure.  ...  A recent work on Face Frontalization using GAN (FFGAN) takes the help of 3D morphable model coefficients to aid with faster training and better convergence [15] . V.  ... 
arXiv:1809.05611v1 fatcat:2xyqoet7ljhhnlueeq265sjxri

Training GANs with predictive projection centripetal acceleration [article]

Li Keke and Zhang Ke and Liu Qiang and Yang Xinmin
2020 arXiv   pre-print
Although remarkable successful in practice, training generative adversarial networks(GANs) is still quite difficult and iteratively prone to cyclic behaviors, as GANs need to solve a non-convex non-concave  ...  Motivated by the ideas of simultaneous centripetal acceleration (SCA) and modified predictive methods (MPM), we propose a novel predictive projection centripetal acceleration (PPCA) methods to alleviate  ...  Both PPCA and APPCA (γ = 1, α = 0.1, β = 0.3) also exponentially converge to the origin and the later seems faster.  ... 
arXiv:2010.03322v2 fatcat:7xqoiznjfbg2tfp26oqvyknjmy

Enforcing constraints for interpolation and extrapolation in Generative Adversarial Networks [article]

Panos Stinis, Tobias Hagge, Alexandre M. Tartakovsky, Enoch Yeung
2018 arXiv   pre-print
Second, the resulting modified interpolator is used for extrapolation where the constraints are enforced after each step through projection on the space of constraints.  ...  Generative Adversarial Networks (GANs) are becoming popular choices for unsupervised learning.  ...  As we have mentioned before, the practice of training GANs has shown that the discriminator can train faster and more effectively than the generator.  ... 
arXiv:1803.08182v1 fatcat:6qjxn63nejf75oigtqwkam7zki

Correlation between surface morphologies and crystallographic structures of GaN layers grown by MOCVD on sapphire

J. L. Rouviere, M. Arlery, R. Niebuhr, K. H. Bachem, Olivier Briot
1996 MRS Internet Journal of Nitride Semiconductor Research  
Convergent Beam Electron Diffraction studies were particularly important to determine the polarity of the GaN layers.  ...  These pyramids are formed when the tiny Ga-polar IDs grow faster than the surrounding N-polar matrix. Flat GaN layers are unipolar, with a Ga polarity.  ...  Kaufmann, the coordinator of this European project. We thank Dr A. Bourret and Dr B. Daudin for helpful discussions.  ... 
doi:10.1557/s1092578300002052 fatcat:qi54zueexvdxhkand4q2aomcle

AutoLoss: Learning Discrete Schedules for Alternate Optimization [article]

Haowen Xu, Hao Zhang, Zhiting Hu, Xiaodan Liang, Ruslan Salakhutdinov, Eric Xing
2018 arXiv   pre-print
Appropriately scheduling the optimization of a task objective or a set of parameters is usually crucial to the quality of convergence.  ...  We apply AutoLoss on four ML tasks: d-ary quadratic regression, classification using a multi-layer perceptron (MLP), image generation using GANs, and multi-task neural machine translation (NMT).  ...  For example, comparing to GAN 1:1, AUTOLOSS improves the converged IS for 0.5, and is almost 3x faster to achieve where GAN 1:1 converges (IS = 8.6) in average.  ... 
arXiv:1810.02442v1 fatcat:hxxbjrl5s5ad5jekolgnrthdle
« Previous Showing results 1 — 15 out of 7,429 results