This shows you the differences between two versions of the page.

Link to this comparison view

Next revision
Previous revision
Last revision Both sides next revision
generative_model [2017/09/11 01:44] external edit
generative_model [2018/12/04 22:08]
Line 303: Line 303:
 https://​arxiv.org/​abs/​1705.09783 Good Semi-supervised Learning that Requires a Bad GAN https://​arxiv.org/​abs/​1705.09783 Good Semi-supervised Learning that Requires a Bad GAN
 +https://​arxiv.org/​pdf/​1710.07035.pdf Generative Adversarial Networks: An Overview
 +https://​openreview.net/​forum?​id=ByQpn1ZA- Many Paths to Equilibrium:​ GANs Do Not Need to Decrease a Divergence At Every Step 
 +https://​arxiv.org/​pdf/​1802.03006.pdf Learning and Querying Fast Generative Models for Reinforcement Learning
 + 1) we
 +provide the first comparison of deterministic and stochastic,
 +pixel-space and state-space models w.r.t. speed and accuracy,
 +applied to challenging environments from the Arcade
 +Learning Environment (ALE, Bellemare et al., 2013); 2) we
 +demonstrate state-of-the-art environment modeling accuracy
 +(as measured by log-likelihoods) with stochastic state-space
 +models that efficiently produce diverse yet consistent rollouts;
 +3) using state-space models, we show model-based RL
 +results on MS PACMAN, and obtain significantly improved
 +performance compared to strong model-free baselines, and
 +4) we show that learning to query the model further increases
 +policy performance
 +https://​arxiv.org/​pdf/​1802.02664.pdf Geometry Score: A Method For Comparing Generative Adversarial Networks
 +One of the biggest challenges in the research of
 +generative adversarial networks (GANs) is assessing
 +the quality of generated samples and detecting
 +various levels of mode collapse. In this work,
 +we construct a novel measure of performance of
 +a GAN by comparing geometrical properties of
 +the underlying data manifold and the generated
 +one, which provides both qualitative and quantitative
 +means for evaluation. Our algorithm can
 +be applied to datasets of an arbitrary nature and
 +is not limited to visual data. We test the obtained
 +metric on various real-life models and datasets
 +and demonstrate that our method provides new
 +insights into properties of GANs.
 +https://​arxiv.org/​abs/​1804.08682 Boltzmann Encoded Adversarial Machines
 +https://​arxiv.org/​abs/​1805.00020 A Guide to Constraining Effective Field Theories with Machine Learning
 +The best results are found for likelihood ratio estimators trained with extra information about the score, the gradient of the log likelihood function with respect to the theory parameters. ​
 +https://​arxiv.org/​pdf/​1805.08318.pdf Self-Attention Generative Adversarial Networks
 +https://​avg.is.tuebingen.mpg.de/​uploads_file/​attachment/​attachment/​424/​Mescheder2018ICML.pdf https://​github.com/​LMescheder/​GAN_stability
 +https://​arxiv.org/​abs/​1807.00374v2 Augmented Cyclic Adversarial Learning for Domain Adaptation
 +https://​arxiv.org/​abs/​1807.03026v1 Pioneer Networks: Progressively Growing Generative Autoencoder
 +we propose the Progressively Growing Generative Autoencoder (PIONEER) network which achieves high-quality reconstruction with 128×128 images without requiring a GAN discriminator
 +https://​arxiv.org/​abs/​1807.04720 The GAN Landscape: Losses, Architectures,​ Regularization,​ and Normalization
 +https://​arxiv.org/​abs/​1807.09295 Improved Training with Curriculum GANs
 +In this paper we introduce Curriculum GANs, a curriculum learning strategy for training Generative Adversarial Networks that increases the strength of the discriminator over the course of training, thereby making the learning task progressively more difficult for the generator. We demonstrate that this strategy is key to obtaining state-of-the-art results in image generation. We also show evidence that this strategy may be broadly applicable to improving GAN training in other data modalities.
 +https://​lilianweng.github.io/​lil-log/​2017/​08/​20/​from-GAN-to-WGAN.html From GAN to WGAN
 +https://​arxiv.org/​abs/​1809.02145v1 GANs beyond divergence minimization
 + These results suggest that GANs do not conform well to the divergence minimization theory and form a much broader range of models than previously assumed.
 +https://​openreview.net/​pdf?​id=B1xsqj09Fm LARGE SCALE GAN TRAINING FOR
 +https://​arxiv.org/​abs/​1810.09136v1 Do Deep Generative Models Know What They Don't Know?
 +https://​ieeexplore.ieee.org/​stamp/​stamp.jsp?​arnumber=8520899 https://​github.com/​ToniCreswell/​InvertingGAN Inverting the Generator of a Generative Adversarial Network
 +https://​arxiv.org/​abs/​1811.03259 Bias and Generalization in Deep Generative Models: An Empirical Study
 +https://​arxiv.org/​abs/​1807.06358v2 IntroVAE: Introspective Variational Autoencoders for Photographic Image Synthesis