Generative adversarial networks (GANs) have been effective for learning generative models for real-world data. However, accompanied with the generative tasks becoming more and more challenging, ...existing GANs (GAN and its variants) tend to suffer from different training problems such as instability and mode collapse. In this paper, we propose a novel GAN framework called evolutionary GANs (E-GANs) for stable GAN training and improved generative performance. Unlike existing GANs, which employ a predefined adversarial objective function alternately training a generator and a discriminator, we evolve a population of generators to play the adversarial game with the discriminator. Different adversarial training objectives are employed as mutation operations and each individual (i.e., generator candidature) are updated based on these mutations. Then, we devise an evaluation mechanism to measure the quality and diversity of generated samples, such that only well-performing generator(s) are preserved and used for further training. In this way, E-GAN overcomes the limitations of an individual adversarial training objective and always preserves the well-performing offspring, contributing to progress in, and the success of GANs. Experiments on several datasets demonstrate that E-GAN achieves convincing generative performance and reduces the training problems inherent in existing GANs.
We investigate the use of parameterized families of information-theoretic measures to generalize the loss functions of generative adversarial networks (GANs) with the objective of improving ...performance. A new generator loss function, least
th-order GAN (L
GAN), is introduced, generalizing the least squares GANs (LSGANs) by using a
th-order absolute error distortion measure with
(which recovers the LSGAN loss function when
). It is shown that minimizing this generalized loss function under an (unconstrained) optimal discriminator is equivalent to minimizing the
th-order Pearson-Vajda divergence. Another novel GAN generator loss function is next proposed in terms of Rényi cross-entropy functionals with order
,
. It is demonstrated that this Rényi-centric generalized loss function, which provably reduces to the original GAN loss function as
, preserves the equilibrium point satisfied by the original GAN based on the Jensen-Rényi divergence, a natural extension of the Jensen-Shannon divergence.
Experimental results indicate that the proposed loss functions, applied to the MNIST and CelebA data sets, under both DCGAN and StyleGAN architectures, confer performance benefits by virtue of the extra degrees of freedom provided by the parameters
and
, respectively. More specifically, experiments show improvements with regard to the quality of the generated images as measured by the Fréchet inception distance score and training stability. While it was applied to GANs in this study, the proposed approach is generic and can be used in other applications of information theory to deep learning, for example, the issues of fairness or privacy in artificial intelligence.
Full text
Available for:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
The Generative Models have gained considerable attention in unsupervised learning via a new and practical framework called Generative Adversarial Networks (GAN) due to their outstanding data ...generation capability. Many GAN models have been proposed, and several practical applications have emerged in various domains of computer vision and machine learning. Despite GANs excellent success, there are still obstacles to stable training. The problems are Nash equilibrium, internal covariate shift, mode collapse, vanishing gradient, and lack of proper evaluation metrics. Therefore, stable training is a crucial issue in different applications for the success of GANs. Herein, we survey several training solutions proposed by different researchers to stabilize GAN training. We discuss (I) the original GAN model and its modified versions, (II) a detailed analysis of various GAN applications in different domains, and (III) a detailed study about the various GAN training obstacles as well as training solutions. Finally, we reveal several issues as well as research outlines to the topic.
Full text
Available for:
IZUM, KILJ, NUK, PILJ, SAZU, UL, UM, UPUK
•A pull-away function is combined to design a new loss function of the generator.•The self-attention module is used in the networks to enhance deep features.•An automatic data filter is established ...to ensure the quality of generated data.
Rolling bearing fault diagnosis is of great significance to the stable operation of rotating machinery systems. However, the fault data collected in practical engineering is seriously imbalanced, which degrades the diagnosis performance. In this paper, a novel data synthesis method called deep feature enhanced generative adversarial network is proposed to improve the performance of imbalanced fault diagnosis. Firstly, to avoid the mode collapse phenomenon and improve the stability of the generative adversarial networks, a pull-away function is integrated to design a new objective function of the generator. Secondly, a self-attention module is utilized in the networks to enhance the deep features of real signals, thereby the quality of synthesized data is improved. Finally, an automatic data filter is established to timely ensure the accuracy and diversity of synthesized samples. Experiments are implemented on two rolling bearing datasets. The results indicate that the proposed method outperforms other intelligent methods and shows great potential in imbalanced fault diagnosis.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP
Image inpainting that completes large free-form missing regions in images is a promising yet challenging task. State-of-the-art approaches have achieved significant progress by taking advantage of ...generative adversarial networks (GAN). However, these approaches can suffer from generating distorted structures and blurry textures in high-resolution images (e.g., <inline-formula><tex-math notation="LaTeX">512\times 512</tex-math> <mml:math><mml:mrow><mml:mn>512</mml:mn><mml:mo>×</mml:mo><mml:mn>512</mml:mn></mml:mrow></mml:math><inline-graphic xlink:href="fu-ieq1-3156949.gif"/> </inline-formula>). The challenges mainly drive from (1) image content reasoning from distant contexts, and (2) fine-grained texture synthesis for a large missing region. To overcome these two challenges, we propose an enhanced GAN-based model, named A ggregated C O ntextual- T ransformation GAN ( AOT-GAN ), for high-resolution image inpainting. Specifically, to enhance context reasoning, we construct the generator of AOT-GAN by stacking multiple layers of a proposed AOT block. The AOT blocks aggregate contextual transformations from various receptive fields, allowing to capture both informative distant image contexts and rich patterns of interest for context reasoning. For improving texture synthesis, we enhance the discriminator of AOT-GAN by training it with a tailored mask-prediction task. Such a training objective forces the discriminator to distinguish the detailed appearances of real and synthesized patches, and in turn facilitates the generator to synthesize clear textures. Extensive comparisons on Places2, the most challenging benchmark with 1.8 million high-resolution images of 365 complex scenes, show that our model outperforms the state-of-the-art. A user study including more than 30 subjects further validates the superiority of AOT-GAN. We further evaluate the proposed AOT-GAN in practical applications, e.g., logo removal, face editing, and object removal. Results show that our model achieves promising completions in the real world. We release codes and models in https://github.com/researchmm/AOT-GAN-for-Inpainting .
•CatGAN and AAE are introduced in unsupervised fault diagnosis of rolling bearings for their great ability of unsupervised clustering and mapping respectively.•By adding a classifier on the latent ...layer of AAE, we propose a new model named CatAAE for unsupervised clustering and exhibit the better performance compared with other methods.•Mixed time-frequency features are employed in the method to get a better robustness under different environments.•Considering about the expenses of labeling data, the proposed unsupervised method is more practical for application.
Fault diagnosis of rolling bearing has been research focus to improve the productivity and guarantee the operation security. In general, traditional approaches need prior knowledge of possible features and a mass of labeled data. Due to the complexity of working conditions, it costs a lot of time to label the monitoring data. In this paper, Categorical Adversarial Autoencoder (CatAAE) is proposed for unsupervised fault diagnosis of rolling bearings. The model trains an autoencoder through an adversarial training process and imposes a prior distribution on the latent coding space. Then a classifier tries to cluster the input examples by balancing mutual information between examples and their predicted categorical class distribution. The latent coding space and training process are presented to investigate the advantage of proposed model. Classic rotating machinery datasets have been employed to testify the effectiveness of the proposed diagnosis method. The experimental results indicate that the proposed method achieved satisfactory performance and high clustering indicators with strong robustness when environmental noise and motor load changed.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UL, UM, UPCLJ, UPUK, ZRSKP
•A fast, generative adversarial network (GAN) based anomaly detection approach.•f−AnoGAN is suitable for real-time anomaly detection applications.•Enables anomaly detection on the image level and ...localization on the pixel level.•Wasserstein GAN (WGAN) training and subsequent encoder training via unsupervised learning on normal data.•Comprehensive experimental evaluation and comparison with alternative approaches.
Obtaining expert labels in clinical imaging is difficult since exhaustive annotation is time-consuming. Furthermore, not all possibly relevant markers may be known and sufficiently well described a priori to even guide annotation. While supervised learning yields good results if expert labeled training data is available, the visual variability, and thus the vocabulary of findings, we can detect and exploit, is limited to the annotated lesions. Here, we present fast AnoGAN (f-AnoGAN), a generative adversarial network (GAN) based unsupervised learning approach capable of identifying anomalous images and image segments, that can serve as imaging biomarker candidates. We build a generative model of healthy training data, and propose and evaluate a fast mapping technique of new data to the GAN’s latent space. The mapping is based on a trained encoder, and anomalies are detected via a combined anomaly score based on the building blocks of the trained model – comprising a discriminator feature residual error and an image reconstruction error. In the experiments on optical coherence tomography data, we compare the proposed method with alternative approaches, and provide comprehensive empirical evidence that f-AnoGAN outperforms alternative approaches and yields high anomaly detection accuracy. In addition, a visual Turing test with two retina experts showed that the generated images are indistinguishable from real normal retinal OCT images. The f-AnoGAN code is available at https://github.com/tSchlegl/f-AnoGAN.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP
A prior-knowledge-guided deep-learning-enabled (PK-DL) synthesis method is proposed for enhancing the transmission bandwidth and phase shift range of metacells used for the design of metalens ...antennas. The algorithm of conditional deep convolutional generative adversarial network (cDCGAN) is utilized in the proposed deep-learning (DL) method. Prior knowledge, including well-known fundamental electromagnetic theorems and experience in antenna design, is purposely applied at the early stage of the proposed method to strategically guide and speed up the synthesis. The proposed intelligent method provides the design of pixelated metacells with high degrees of freedom so that the key performance of the synthesized metacells exceeds the existing limit of conventional design methods by generating a rich profusion of cell patterns. For example, the synthesized triple-layer metacell achieves the −1 dB phase shift range of 330° breaking the limit of 308° derived by existing techniques. The proposed synthesis method also provides the additional capability to flexibly control the phase shift not only at the center frequency but also over a frequency range of interest. A Ku-band metalens antenna formed with the synthesized metacells demonstrates the achieved 1 and 3 dB gain bandwidths increase by 52.2% and 42.6%, respectively, compared to the metalens antenna using the well-known Jerusalem cross (JC) metacells. The proposed method extends the capability for the synthesis of metacells and metalens antennas with enhanced performance.