Ultrasound Nerve Segmentation Using Deep Probabilistic Programming

In this paper, a generative adversarial network based on SMPL model is used to reconstruct three-dimensional (3D) human mesh from a single image. The generative adversarial network consists of two parts (generative network and adversarial network). The generative network is made up of a deep residual network and a multilayer perception network. The goal is to generate suitable SMPL model parameters and camera parameters which fit the 2D and 3D position annotation of human skeleton joints. The adversarial network judges whether the shape and pose parameters generated by the generative network match the real human body through a discriminator. The purpose is to guide the generative network output parameters that conform to the laws of human kinematics. A geometric prior of the 3D joint angle limitation of the human body is introduced into the network to further constrain the generated human pose. Experiments verify the effectiveness of the algorithm.

Probabilistic programming refers to the idea of using standard programming constructs for specifying probabilistic models from machine learning and statistics, and employing generic inference algorithms for answering various queries on these models, such as posterior inference and estimation of model evidence. Although this idea itself is not new and was, in fact, explored by several programming-language and statistics researchers in the early 2000, it is only in the last few years that probabilistic programming has gained a large amount of attention among researchers in machine learning and programming languages, and that expressive and efficient probabilistic programming systems (such as Anglican, Church, Figaro, Infer.net, PSI, PyMC, Stan, and Venture) started to appear and have been taken up by a nontrivial number of users. The primary goal of my talk is to introduce probabilistic programming to the CONCUR/QUEST/FORMATS audience. At the end of my talk, I want the audience to understand basic results and techniques in probabilistic programming and to feel that these results and techniques are relevant or at least related to what she or he studies, although they typically come from foreign research areas, such as machine learning and statistics. My talk will contain both technical materials and lessons that I learnt from my machine-learning colleagues in Oxford, who are developing a highly-expressive higher-order probabilistic programming language, called Anglican. It will also include my work on the denotational semantics of higher-order probabilistic programming languages and their inference algorithms, which are jointly pursued with colleagues in Cambridge, Edinburgh, Oxford and Tubingen.

Flooding is one of the most frequent natural hazards and causes more economic loss than all the other natural hazards. Fast and accurate flood prediction has significance in preserving lives, minimizing economic damage, and reducing public health risks. However, current methods cannot achieve speed and accuracy simultaneously. Numerical methods can provide high-fidelity results, but they are time-consuming, particularly when pursuing high accuracy. Conversely, neural networks can provide results in a matter of seconds, but they have shown low accuracy in flood map generation by all existing methods. This work combines the strengths of numerical methods and neural networks and builds a framework that can quickly and accurately model the high-fidelity flood inundation map with detailed water depth information. In this paper, we employ the U-Net and generative adversarial network (GAN) models to recover the lost physics and information from ultra-fast, low-resolution numerical simulations, ultimately presenting high-resolution, high-fidelity flood maps as the end results. In this study, both the U-Net and GAN models have proven their ability to reduce the computation time for generating high-fidelity results, reducing it from 7–8 h down to 1 min. Furthermore, the accuracy of both models is notably high.

This paper introduces a novel automotive radar interference mitigation approach using a generative adversarial network (GAN). Instead of tackling the mutual interference in the time domain, a generative adversarial network is trained and used to recover the complex signal in the frequency domain, namely on the complex range profile obtained after the fast Fourier transform of fast-time samples (RFFT spectrum). It is shown that by employing the gated convolution and an attention mechanism, the generator network has the ability to learn the amplitude and phase information for missing data from the remaining signal. Experimental results show that the proposed method can provide a remarkable improvement in signal-to-interference-plus-noise ratio (SINR) and preserves its robustness in severely disturbed scenarios that are much more complex than the training dataset.

The terahertz (THz) metamaterial sensor design is typically complex and requires substantial expertise in physics. To simplify this process, we propose a novel reverse design model based on an improved conditional generative adversarial network that integrates self-attention generative adversarial network and Wasserstein generative adversarial network (WGAN) networks, and is referred to as the self-attention conditional Wasserstein GAN (SACW-GAN) model. By using the target response of the sensor as the input to the generator network, and incorporating labeling information, an attention mechanism, and the Wasserstein distance, we achieve effective reverse design of THz metamaterial sensors. The simulation results demonstrate the model's high performance, with spectral and image accuracies of 95% and 97%, respectively. This deep learning approach offers new perspectives and methodologies for the reverse design and application of THz metamaterial sensors, significantly advancing the field.

Generative adversarial networks (GANs) with gaming abilities have been widely applied in image generation. However, gamistic generators and discriminators may reduce the robustness of the obtained GANs in image generation under varying scenes. Enhancing the relation of hierarchical information in a generation network and enlarging differences of different network architectures can facilitate more structural information to improve the generation effect for image generation. In this paper, we propose an enhanced GAN via improving a generator for image generation (EIGGAN). EIGGAN applies a spatial attention to a generator to extract salient information to enhance the truthfulness of the generated images. Taking into relation the context account, parallel residual operations are fused into a generation network to extract more structural information from the different layers. Finally, a mixed loss function in a GAN is exploited to make a tradeoff between speed and accuracy to generate more realistic images. Experimental results show that the proposed method is superior to popular methods, i.e., Wasserstein GAN with gradient penalty (WGAN-GP) in terms of many indexes, i.e., Frechet Inception Distance, Learned Perceptual Image Patch Similarity, Multi-Scale Structural Similarity Index Measure, Kernel Inception Distance, Number of Statistically-Different Bins, Inception Score and some visual images for image generation.

Image-to-image (I2I) translation using generative adversarial networks (GANs) has become a standard approach across numerous scientific domains. In solar physics, GANs have become a popular approach to reconstructing unavailable modalities from physically related modalities that are available at the time of interest. However, the scientific validity of outputs generated by GANs has been largely overlooked thus far. We address a critical challenge in generative deep learning models, namely, their tendency to produce visually and statistically convincing outputs that might be physically inconsistent with the input data. We measured the discrepancy between GAN-generated solar images and real observations in two applications: the generation of chromospheric images from photospheric images and the generation of magnetograms from extreme ultraviolet observations. In each case, we considered both global and application-specific performance metrics. Next, we investigated non-adversarial training strategies and network architectures, whose behavior could be adapted to the input at hand. Specifically, we propose an architecture that modulates the generative model's internal feature maps with input-related information, thereby favoring the transfer of input-output mutual information to the output. Global metrics show that GANs consistently fall short of non-adversarial U-net translation models in physics-constrained applications due to the generation of visually appealing features that do not have any real physical correspondence. Such features are referred to as hallucinations. Additional conditioning procedures carried out via the U-net model, based on the modulation of internal feature maps, can significantly enhance cross-modal image-to-image translation. Our work demonstrates that adaptive instance modulation results in reconstructions that are less prone to hallucinations compared to adversarial settings. An increased robustness to hallucinations is an important advantage in solar physics research where spurious features can be highly problematic.

Generative adversarial network (GAN) has been applied for low-dose CT images to predict normal-dose CT images. However, the undesired artifacts and details bring uncertainty to the clinical diagnosis. In order to improve the visual quality while suppressing the noise, in this paper, we mainly studied the two key components of deep learning based low-dose CT (LDCT) restoration models—network architecture and adversarial loss, and proposed a disentangled noise suppression method based on GAN (DNSGAN) for LDCT. Specifically, a generator network, which contains the noise suppression and structure recovery modules, is proposed. Furthermore, a multi-scaled relativistic adversarial loss is introduced to preserve the finer structures of generated images. Experiments on simulated and real LDCT datasets show that the proposed method can effectively remove noise while recovering finer details and provide better visual perception than other state-of-the-art methods.

Segmentation of breast cancer tumor plays an important role in identifying the location of the tumor, to know the shape of tumor and hence the stage of breast cancer. This paper deals with the segmentation of tumor from whole mammographic mass images using Generative Adversarial Network (GAN). A mini dataset was considered with mammograms and their corresponding ground truth images. Pre-processing like image format conversion, enhancement, pectoral muscle removal and resizing was performed on raw mammogram images. GANs have two neural nets called generative and discriminative networks that compete against each other to obtain the segmentation output. PIX2PIX is a conditional GAN variant which has U-Net as the Generator network and a simple deep neural net as the discriminator. The input to the network was pair of pre-processed mass image and the associated ground truth. A binary image with highlighted tumor was obtained as output. The performance of GAN was evaluated by plotting Generator and discriminator loss. The segmented output was compared with corresponding ground truth. Metrics like Jaccard index, Jaccard distance and Dice-coefficient were calculated. A Dice-coefficient and Jaccard index of 90% and 88.38% was achieved. In future, higher accuracy could be achieved by involving larger dataset to make the system robust.

Many machine learning models require huge datasets to get trained, to make correct predictions, or to increase their accuracy, that's where generative adversarial network (GAN) comes into the picture. GAN is a deep learning model, which can be used to create artificial data. It consists of two neural networks that cooperate in a way that resembles a game: a generator network and a discriminator network. Within the suggested model in this work, a dense motion network-based first-order motion model for image animation is developed. Here a trained GAN extracts the face landmarks from the driving video and develops the embedding model to create the synthesis video using the dedicated module to prepare the Deepfakes, employing key point detectors as a baseline. Lastly, an approach that makes use of dense motion networks to increase the effectiveness of a collection of GAN generators is provided. With the help of the sequel driving combination of driving video with the source image, the given results produce the augmented animation video. Hence, it is tried to implement a lighter model which consists of modules, with the approximate same accuracy when compared with other implementations of GAN. This work has a wide range of applications, including doubling dataset counts with a small number of sources, creating real-time backgrounds and characters for the gaming and animation industries using CG platforms, translating clothes, predicting videos, creating 3D objects, etc.

Recovering images from highly undersampled measurements has a wide range of applications across imaging sciences. State-of-the-art analytics however are not aware of the image perceptual quality, and demand iterative algorithms that incur significant computational overhead. To sidestep these hurdles, this paper brings forth a novel compressive imaging framework using deep neural networks that approximates a low-dimensional manifold of images using generative adversarial networks. To ensure the images are consistent with the measurements a recurrent GAN (RGAN) architecture is deployed that consists of multiple alternative blocks of generator networks and affine projection, which is then followed by a discriminator network to score the perceptual quality of the generated images. A deep residual network with skip connections is used for the generator, while the discriminator is a multilayer perceptron. Experiments performed with real-world contrast enhanced MRI data corroborate the diagnostic quality of the retrieved images relative to state-of-the-art CS schemes. In addition, it achieves about two-orders of magnitude faster reconstruction.

The amount of structural damage image data produced in the aftermath of an earthquake can be staggering. It is challenging for a few human volunteers to efficiently filter and tag these images with meaningful damage information. There are several solution to automate post-earthquake reconnaissance image tagging using Machine Learning (ML) solutions to classify each occurrence of damage per building material and structural member type. ML algorithms are data driven; improving with increased training data. Thanks to the vast amount of data available and advances in computer architectures, ML and in particular Deep Learning (DL) has become one of the most popular image classification algorithms producing results comparable to and in some cases superior to human experts. These kind of algorithms need the input images used for the training to be labeled, and even if there is a large amount of images most of them are not labeled and it takes structural engineers a large amount of time to do it. The current data earthquakes image data bases do not contain the label information or is incomplete slowing significantly the advance of a solution and are incredible difficult to search. To be able to train a ML algorithm to classify one of the structural damages it took the architecture school an entire year to gather 200 images of the specific damage. That number is clearly not enough to avoid overfitting so for this thesis we decided to generate synthetic images for the specific structural damage. In particular we attempt to use Generative Adversarial Neural Networks (GANs) to generate the synthetic images and enable the fast classification of rail and road damage caused by earthquakes. Fast classification of rail and road damage can allow for the safety of people and to better prepare the reconnaissance teams that manage recovery tasks. GANs combine classification neural networks with generative neural networks. For this thesis we will be combining a convolutional neural network (CNN) with a generative neural network. By taking a classifier trained in a GAN and modifying it to classify other images the classifier can take advantage of the GAN training without having to find more training data. The classifier trained in this way was able to achieve an 88\% accuracy score when classifying images of structural damage caused by earthquakes.

The deep learning models play an essential role in many areas, including medical image analysis. These models extract important features without human intervention. In this paper, we propose a deep convolution neural network, named as deep U-Net model, for the segmentation of the cell nucleus, a critical functional unit that determines the function and structure of the body. The nucleus contains all kinds of DNA, RNA, chromosomes, and genes governing all life activities, and its disorder may lead to different types of diseases such as cancer, heart disease, diabetes, Alzheimer’s, etc. If the nucleus structure is known correctly, diseases due to nucleus disorder may be detected early. It may also reduce the drug discovery time if the shape and size of the nucleus are known. We evaluate the performance of the proposed models on the nucleus segmentation data set used by the Data Science Bowl 2018 competition hosted by Kaggle. We compare its performance with that of the U-Net, Attention U-Net, R2U-Net, Attention R2U-Net, and both versions of the U-Net++ with and without supervision, in terms of loss, dice coefficient, dice loss, intersection over union, and accuracy. Our model performs better than the existing models.

Gas-liquid two-phase flow is widely involved in many scientific and technological fields, such as energy, electricity, nuclear energy, aerospace and environmental protection. In some fields, extracting the accurate position of bubbles in space can not only accurately capture the characteristics of bubbles in two-phase flow, but also plays an important role in the subsequent research like bubble tracking. It has got some progresses to use Convolutional Neural Network (CNNs) to identify bubbles in gas-liquid two-phase flow, while accurate pixel segmentation map in the bubble identification problem is more desirable in many areas. In this paper, VGG16-FCN model and U-Net model are utilized to identify bubbles in two-phase flow images from the perspective of semantic segmentation. LabelMe is used to label the images in the experiment, which can remove the noise in the original image. In addition, bubble pixels with low ratio relative to the background affects the loss function value tinily which cause the irrational evaluation for the recognition in traditional semantic segmentation, thus, Dice loss is used as the loss function for training to improve the recognition effect. The research results show that the two deep learning models have strong feature extraction ability and accurately detect the bubble boundary.

Generative Adversarial Network is the topic of interest in today’s research in the field of image processing and computer vision. A basic GAN model was introduced by Ian Goodfellow et al. in 2014. After that advancement in the field of research in GAN models has been application specific. In computer vision and image to image translation GANs are playing very effective role either in the case of face detection and recognition or in image resolution enhancement and image augmentation. This paper represents a concise overview of various GAN models along with their features and applications. Pix2Pix and conditional GAN models work upon paired datasets while other models like cycle GAN, discover GAN, dual GAN, info GAN, deep convolutional GAN etc. work upon unpaired datasets. Various image datasets which are commonly used for training of generator and discriminator networks are also discussed in this paper. Since partial mode collapse is a common problem to occur during training process for all models, therefore various normalization techniques are also preferred during the training of generator and discriminator networks. As the advancements in GAN models are increasing at a very fast rate, soon these models will also be preferred in commercial applications.