Radiography is an imaging technique used in a variety of applications, such as medical diagnosis, airport security, and nondestructive testing. We present a deep learning system for extracting information from radiographic images. We perform various prediction tasks using our system, including material classification and regression on the dimensions of a given object that is being radiographed. Our system is designed to address the sparse-data issue for radiographic nondestructive testing applications. It uses a radiographic simulation tool for synthetic data augmentation, and it uses transfer learning with a pre-trained convolutional neural network model. Using this system, our preliminary results indicate that the object geometry regression task saw an improvement of 70% in the R-squared value when using a multi-regime model. In addition, we increase the performance of the object material classification tasks by utilizing data from different imaging systems. In particular, using neutron imaging improved the material classification accuracy by 20% when compared to x-ray imaging.

In this paper, we address the challenge of analyzing simulation data on HPC systems by using Apache Spark, which is a Big Data framework. One of the main problems we encountered with using Spark on HPC systems is the ephemeral data explosion, which is brought about by the curse of persistence in the Spark framework. Data persistence is essential in reducing I/O, but it comes at the cost of storage space. We show that in some cases, Spark scratch data can consume an order of magnitude more space than the input data being analyzed, leading to fatal out-of-disk errors. We investigate the real-world application of scaling machine learning algorithms to predict and analyze failures in multi-physics simulations on 76TB of data (over one trillion training examples). This problem is 2--3 orders of magnitude larger than prior work. Based on extensive experiments at scale, we provide several concrete recommendations as state-of-the-practice, and demonstrate a 7x reduction in disk utilization with negligible increases or even decreases in runtime.

Purpose: To identify a subgroup at high risk for loco-regional recurrence (LRR) from T1-2 breast cancer with negative lymph nodes (N0) after mastectomy by using a meta-analysis. Methods and materials: Published studies on the relationship between clinical features and LRR of breast cancer were identified from public databases, including PubMed, EMBASE, and the Cochrane Library. High-risk features for LRR in this patient population were defined based on the pooled results of meta-analysis. Results: For the meta-analysis, a total of 11244 breast cancers with pT1-2N0 after mastectomy from 20 publications were included for analysis. The pooled results indicated that age (hazard ratio (HR) 1.77, P=0.001), lymphovascular invasion (LVI) (HR 2.23, P<0.001), histologic grade (HR 1.66, P<0.001), HER2 status (HR 1.65, P=0.027), menopausal status (HR 1.36, P=0.015), and surgical margins (HR 2.56, P=0.014) were associated with a significantly increased risk of developing LRR in this patient population group, but not for tumor size (HR 1.32, P=0.23), systematic therapy (HR 1.67, P=0.20), and hormonal receptor status (HR 1.04, P=0.73). Conclusion: In the current study, patients with young age, positive LVI, high histologic grade, HER-2 positive, premenopausal, and positive surgical margins have an increased risk of developing LRR. Further prospective trials are needed to clearly define the role of adjuvant postmastectomy radiotherapy in T1-2N0 breast cancer at high risk of developing LRR.

The Arbitrary Lagrangian-Eulerian (ALE) method is used in a variety of engineering and scientific applications for enabling multi-physics simulations. Unfortunately, the ALE method can suffer from failures that require users to adjust a set of parameters to control mesh relaxation. In this paper, we present a deep learning framework for predicting mesh relaxation in ALE simulations. Our framework is designed to train a neural network using data generated from existing ALE simulations developed by expert users. In order to capture the spatial coherence inherent in simulations, we apply convolutional-deconvolutional neural networks to achieve up to 0.99 F1 score in predicting mesh relaxation.

Machine learning (ML) is being applied in a number of everyday contexts from image recognition, to natural language processing, to autonomous vehicles, to product recommendation. In the science realm, ML is being used for medical diagnosis, new materials development, smart agriculture, DNA classification, and many others. In this article, we describe the opportunities of using ML in the area of scientific workflow management. Scientific workflows are key to today’s computational science, enabling the definition and execution of complex applications in heterogeneous and often distributed environments. We describe the challenges of composing and executing scientific workflows and identify opportunities for applying ML techniques to meet these challenges by enhancing the current workflow management system capabilities. We foresee that as the ML field progresses, the automation provided by workflow management systems will greatly increase and result in significant improvements in scientific productivity.

The convergence of high-performance computing (HPC) and Big Data is a necessity with the push towards extreme-scale computing. As HPC simulations become more complex, the analytics need to process larger amounts of data, which poses significant challenges for coupling HPC simulations with Big Data analytics. This poster presents a novel node-local approach that uses a workflow management system (WMS) to enable the coupling between the simulations and the analytics in scientific workflows by leveraging node-local non-volatile random-access memory (NVRAM).

Recurrent miscarriage is defined as the loss of 3 or more consecutive pregnancies; however, the underlying immunologic mechanisms that trigger pregnancy loss remain largely unelucidated. Galectin-9 (Gal-9) may modulate a variety of biologic functions and play an important role in Th1/Th2 immune deviation. To analyze the mechanism of Gal-9 in abortion, we used the classical abortion-prone mouse model (DBA/2-mated CBA/J mice) to detect the expression of Gal-9 at the maternal-fetal interface. We also mimicked the immune environment of pregnancy by culturing trophoblast cells with peripheral blood mononuclear cells (PBMCs) to explore how Gal-9 might be involved in the pathogenesis of abortion. We found that the expression levels of Gal-9 in abortion-prone matings were lower than that for controls. Using a coculture system, we detected a Th1 preponderance in the coculture from abortion-prone matings. Furthermore, Gal-9 blockade augmented the imbalance of Th1/Th2 immunity in abortion-prone matings by promoting the secretion of Th1-derived cytokines in coculture, while there was a Th2 preponderance when we administered recombinant Gal-9. In conclusion, our results suggest that the Gal-9 signal is important for the regulation of PBMC function toward a Th2 bias at the maternal-fetal interface, which is beneficial for the maintenance of a normal pregnancy.

Automation of the execution of computational tasks is at the heart of improving scientific productivity. Over the last years, scientific workflows have been established as an important abstraction that captures data processing and computation of large and complex scientific applications. By allowing scientists to model and express entire data processing steps and their dependencies, workflow management systems relieve scientists from the details of an application and manage its execution on a computational infrastructure. As the resource requirements of today’s computational and data science applications that process vast amounts of data keep increasing, there is a compelling case for a new generation of advances in high-performance computing, commonly termed as extreme-scale computing, which will bring forth multiple challenges for the design of workflow applications and management systems. This paper presents a novel characterization of workflow management systems using features commonly associated with extreme-scale computing applications. We classify 15 popular workflow management systems in terms of workflow execution models, heterogeneous computing environments, and data access methods. The paper also surveys workflow applications and identifies gaps for future research on the road to extreme-scale workflows and management systems.

The Arbitrary Lagrangian-Eulerian (ALE) method is used in a variety of engineering and scientific applications for enabling multi-physics simulations. Unfortunately, the ALE method can suffer from simulation failures that require users to adjust parameters iteratively in order to complete a simulation. In this paper, we present a supervised learning framework for predicting conditions leading to simulation failures. To our knowledge, this is the first time machine learning has been applied to ALE simulations. We propose a novel learning representation for mapping the ALE domain onto a supervised learning formulation. We analyze the predictability of these failures and evaluate our framework using well-known test problems.

One of the main difficulties when running Arbitrary Lagrangian-Eulerian (ALE) simulations is determining how much to relax the mesh during the Eulerian step. This determination is currently made by the user on a simulation-by-simulation basis. We present a Learning Algorithm-Generated Empirical Relaxer (LAGER) which uses a regressive random forest algorithm to automate this decision process. We also demonstrate that LAGER successfully relaxes a variety of test problems, maintains simulation accuracy, and has the potential to significantly decrease both the person-hours and computational hours needed to run a successful ALE simulation.

Streamline tracing is an important tool used in many scientific domains for visualizing and analyzing flow fields. In this work, we examine a shared memory multi-threaded approach to streamline tracing that targets emerging data-intensive architectures. We take an in-depth look at data management strategies for streamline tracing in terms of issues, such as memory latency, bandwidth, and capacity limitations, that are applicable to future HPC platforms. We present two data management strategies for streamline tracing and evaluate their effectiveness for data-intensive architectures with locally attached Flash. We provide a comprehensive evaluation of both strategies by examining the strong and weak scaling implications of a variety of parameters. We also characterize the relationship between I/O concurrency and I/O efficiency to guide the selection of strategy based on use case. From our experiments, we find that using kernel-managed memory-map for out-of-core streamline tracing can outperform optimized user-managed cache.

Derived field generation is a critical aspect of many visualization and analysis systems. This capability is frequently implemented by providing users with a language to create new fields and then translating their "programs" into a pipeline of filters that are combined in sequential fashion. Although this design is highly extensible and practical for development, the runtime characteristics of the typical implementation are poor, since it iterates over large arrays many times. As we reconsider visualization and analysis systems for many-core architectures, we must re-think the best way to implement derived fields while being cognizant of data movement. In this paper, we describe a flexible Python-based framework that realizes efficient derived field generation on many-core architectures using OpenCL. Our framework supports the development of different execution strategies for composing operations using a common library of building blocks. We present an evaluation of our framework by testing three execution strategies to explore tradeoffs between runtime performance and memory constraints. We successfully demonstrate our framework in an HPC environment using the vortex detection application on a large-scale simulation.

This article presents a simple framework for progressive processing of high-resolution images with minimal resources. We demonstrate this framework's effectiveness by implementing an adaptive, multi-resolution solver for gradient-based image processing that, for the first time, is capable of handling gigapixel imagery in real time. With our system, artists can use commodity hardware to interactively edit massive imagery and apply complex operators, such as seamless cloning, panorama stitching, and tone mapping. We introduce a progressive Poisson solver that processes images in a purely coarse-to-fine manner, providing near instantaneous global approximations for interactive display (see Figure 1). We also allow for data-driven adaptive refinements to locally emulate the effects of a global solution. These techniques, combined with a fast, cache-friendly data access mechanism, allow the user to interactively explore and edit massive imagery, with the illusion of having a full solution at hand. In particular, we demonstrate the interactive modification of gigapixel panoramas that previously required extensive offline processing. Even with massive satellite images surpassing a hundred gigapixels in size, we enable repeated interactive editing in a dynamically changing environment. Images at these scales are significantly beyond the purview of previous methods yet are processed interactively using our techniques. Finally our system provides a robust and scalable out-of-core solver that consistently offers high-quality solutions while maintaining strict control over system resources.

Many cell surface receptors are multimeric proteins, composed of several structural domains, some involved in ligand recognition, whereas others are responsible for signal transduction. In most cases, the mechanism of how ligand interaction in the extracellular domains leads to the activation of effector domains remains largely unknown. Here we examined how the extracellular ligand binding to the venus flytrap (VFT) domains of the dimeric metabotropic glutamate receptors activate the seven transmembrane (7TM) domains responsible for G protein activation. These two domains are interconnected by a cysteine-rich domain (CRD). We show that any of the four disulfide bridges of the CRD are required for the allosteric coupling between the VFT and the 7TM domains. More importantly, we show that a specific association of the two CRDs corresponds to the active state of the receptor. Indeed, a specific crosslinking of the CRDs with intersubunit disulfide bridges leads to fully constitutively active receptors, no longer activated by agonists nor by allosteric modulators. These data demonstrate that intersubunit movement at the level of the CRDs represents a key step in metabotropic glutamate receptor activation.

Realistic engineering simulation data often have features that are not optimally resolved due to practical limitations on mesh resolution. To be useful to application engineers, vortex characterization techniques must be sufficiently robust to handle realistic data with complex vortex topologies. In this paper, we present enhancements to the vortex topology identification component of an existing vortex characterization algorithm. The modified techniques are demonstrated by application to three realistic data sets that illustrate the strengths and weaknesses of our approach.

In order to understand complex vortical flows in large data sets, we must be able to detect and visualize vortices in an automated fashion. In this paper, we present a feature-based vortex detection and visualization technique that is appropriate for large computational fluid dynamics data sets computed on unstructured meshes. In particular, we focus on the application of this technique to visualization of the flow over a serrated wing and the flow field around a spinning missile with dithering canards. We have developed a core line extraction technique based on the observation that vortex cores coincide with local extrema in certain scalar fields. We also have developed a novel technique to handle complex vortex topology that is based on k-means clustering. These techniques facilitate visualization of vortices in simulation data that may not be optimally resolved or sampled. Results are included that highlight the strengths and weaknesses of our approach. We conclude by describing how our approach can be improved to enhance robustness and expand its range of applicability.

In order to understand complex vortical flows in large data sets, we must be able to detect and visualize vortices in an automated fashion. Additionally, an informative vortex visualization should provide more information that just the location of the vortex core. In this paper, we describe a multistage vortex visualization algorithm that attempts to exploit the availability of additional information to convey vortex characteristics such as core position, extent, strength, etc. The results produced by a traditional vortex detection algorithm are enhanced using a set of postprocessing tools. The multistage visualization method can use the results of any field-based vortex detection technique in which the scalar field has an extrema in the vortex core or can use the results of any method which produces a continuous vortex core line. This flexibility allows the user to select a vortex detection algorithm appropriate for the problem at hand. Our method works on unstructured meshes as well as structured grids and avoids the use of the computationally expensive and potentially nonsmooth gradient during core line extraction and characterization. Its ability to discern both weak and strong vortices as well as vortices in close spatial proximity to each other is another advantage which comes from the use of the local extrema method. Further, our method obtains subcell resolution of the vortex core line through the use of function fitting. We demonstrate the efficacy of our approach by applying it to several test cases. These test cases highlight the strengths and weaknesses of the method. We describe how the algorithm can be improved and draw conclusions based on the results presented herein.

A vortex is characterized by the swirling motion of fluid around a central region. This characterization stems from the visual perception of swirling phenomena that are pervasive throughout the natural world. However, translating this intuitive description of a vortex into a formal definition has been quite a challenge. Despite the lack of a formal definition, various detection algorithms have been implemented that can adequately identify vortices in most computational datasets. This chapter presents an overview of existing detection methods; in particular, the focus is on nine methods that are representative of the state of the art. The chapter begins by presenting three taxonomies for classifying these nine detection methods. It then describes each algorithm, along with pseudo-code where appropriate. Next, the chapter describes a recently developed verification algorithm for swirling flows. The chapter also discusses the different visualization techniques for vortices.

Our main goal in developing visualization techniques for the Hurricane Isabel dataset is to engender better understanding of the underlying physical phenomenon. We want the visualization to produce novel insights into how a hurricane behaves across time, and how the various properties of the hurricane interact with each other. Most importantly, the visualization should tell us which parts of the hurricane are the most destructive, and when and where those destructive parts will reach Florida. In this paper, we describe the various visualization techniques that we applied to the Hurricane Isabel dataset. Although we did not develop any new visualization techniques per se, we were able to effectively utilize existing ones in meaningful ways. Our primary contribution is adapting a set of existing algorithms and devising an overall framework for applying them to the dataset. In particular, we applied several vortex detection algorithms to the time varying dataset in order to automatically identify the various parts of the hurricane. We then developed a framework to automatically track the position of the hurricane across time, while computing both its translational velocity and maximum wind speed. Finally, we experimented with several volume rendering techniques to better understand some of the physical properties of the dataset.