Final Report for "Convenient Multi-Architecture High-Performance Computing"
Explore the source record for details and available documents.
SEARCH · Search NASA
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Deterministic solutions to the Sn radiation transport equation can be computationally expensive to calculate. Reduced-order modeling enables efficient approximation of the full-order model (FOM) solution. We propose a novel method for constructing reduced-order models (ROMs) of the S n radiation transport equation, offline maximizing minimally invasive (OMMI) proper orthogonal decomposition (POD). POD uses the method of snapshots to create a reduced-order basis for constructing an ROM. Minimally invasive POD leverages the sweep infrastructure existing in deterministic transport codes to create a POD-based ROM, even when infeasible by traditional methods. Offline maximizing minimally invasive proper orthogonal decomposition (OMMI-POD) extends minimally invasive POD by performing sweeps offline, therefore maximizing the potential speedup. OMMI-POD does so by creating a library of reduced systems from a training set. This library of reduced systems is then interpolated to provide a rapid approximate solution of the S n radiation transport equation. The model is evaluated on a set of test problems, achieving a low error with a 466 times speedup over the FOM. Also presented is a study of the effect of sampling method on the performance of OMMI-POD, specifically comparing naive uniform sampling to the more accurate and computationally expensive greedy sampling.
Despite great progress in simulating multiphysics problems using the numerical discretization of partial differential equations (PDEs), one still cannot seamlessly incorporate noisy data into existing algorithms, mesh generation remains complex, and high-dimensional problems governed by parameterized PDEs cannot be tackled. Moreover, solving inverse problems with hidden physics is often prohibitively expensive and requires different formulations and elaborate computer codes. Machine learning has emerged as a promising alternative, but training deep neural networks requires big data, not always available for scientific problems. Instead, such networks can be trained from additional information obtained by enforcing the physical laws (for example, at random points in the continuous space-time domain). Such physics-informed learning integrates (noisy) data and mathematical models, and implements them through neural networks or other kernel-based regression networks. Moreover, it may be possible to design specialized network architectures that automatically satisfy some of the physical invariants for better accuracy, faster training and improved generalization. Furthermore, we review some of the prevailing trends in embedding physics into machine learning, present some of the current capabilities and limitations and discuss diverse applications of physics-informed learning both for forward and inverse problems, including discovering hidden physics and tackling high-dimensional problems.
NREL's computational sciences center hosts the largest high performance computing (HPC) capabilities dedicated to energy research while functioning as a living laboratory for energy-efficient computing. NREL's HPC capabilities support the research needs of the Department of Energy's Office of Energy Efficiency and Renewable Energy (EERE). In ten years of operation, HPC use in EERE-sponsored research has grown by a factor of 30, including work in electricity generation, energy efficiency, transportation, and energy system modeling. This paper analyzes this research portfolio, providing examples of individual use cases. The paper documents NREL's history of operating one of the world's most energy-efficient data centers while examining pathways to reduce economic and environmental impact beyond reduction of Power Usage Efficiency (PUE). This paper concludes by examining the unique opportunities created for accelerating improvements in data center efficiency created by combining an HPC system dedicated to energy research and a research program in energy-efficient computing.
Complex multiphysics simulations such as nuclear fuel performance analysis are composed of many submodels used to describe specific phenomena. These phenomena include, as examples, the relationship between stress and strain, heat transfer across a gas gap, and mechanical contact. These submodels work in concert to simulate real-world events, like the behavior of a fuel rod in a reactor. If a simulation tool is able to represent real-world behavior, the tool is said to be validated. While much emphasis is rightly placed on validation, model verification may be undervalued. Verification involves showing that a model performs as intended, that it computes results consistent with its mathematical description. This paper explains the differences between verification and validation and shows how validation should be preceded by verification. Specific verification problems, including several specific to nuclear fuel analysis, are given. Validation results are also presented.
Large, complex scientific and engineering application code have a significant investment in computational kernels to implement their mathematical models. Porting these computational kernels to the collection of modern manycore accelerator devices is a major challenge in that these devices have diverse programming models, application programming interfaces (APIs), and performance requirements. The Kokkos Array programming model provides library-based approach to implement computational kernels that are performance-portable to CPU-multicore and GPGPU accelerator devices. This programming model is based upon three fundamental concepts: (1) manycore compute devices each with its own memory space, (2) data parallel kernels and (3) multidimensional arrays. Kernel execution performance is, especially for NVIDIA® devices, extremely dependent on data access patterns. Optimal data access pattern can be different for different manycore devices – potentially leading to different implementations of computational kernels specialized for different devices. The Kokkos Array programming model supports performance-portable kernels by (1) separating data access patterns from computational kernels through a multidimensional array API and (2) introduce device-specific data access mappings when a kernel is compiled. An implementation of Kokkos Array is available through Trilinos [Trilinos website, http://trilinos.sandia.gov/, August 2011].
Iso-geometric analysis can be viewed as a generalisation of the finite element method. It permits the exact representation of a wider range of geometries including conic sections. This is possible due to the use of concepts employed in computer-aided design. The underlying mathematical representations from computer-aided design are used to capture both the geometry and approximate the solution. In this paper the neutron diffusion equation is solved using iso-geometric analysis. The practical advantages are highlighted by looking at the problem of a circular fuel pin in a square moderator. For this problem the finite element method requires the geometry to be approximated. This leads to errors in the shape and size of the interface between the fuel and the moderator. In contrast to this iso-geometric analysis allows the interface to be represented exactly. It is found that, due to a cancellation of errors, the finite element method converges more quickly than iso-geometric analysis for this problem. A fuel pin in a vacuum was then considered as this problem is highly sensitive to the leakage across the interface. In this case iso-geometric analysis greatly outperforms the finite element method. Due to the improvement in the representation of the geometry iso-geometric analysis can outperform traditional finite element methods. It is proposed that the use of iso-geometric analysis on neutron transport problems will allow deterministic solutions to be obtained for exact geometries. Something that is only currently possible with Monte Carlo techniques. (authors)
DC is launching a new program to recognize noteworthy achievements made by users using High Performance Computing (HPC) technologies. The program’s main goals are to: • Showcase success stories involving HPC in science and industry • Help other users better understand the benefits of adopting HPC and justify HPC investments, especially for SMBs • Demonstrate the value of HPC to funding bodies • Expand public support for increased HPC investments.
Compact data representations in quantum systems are crucial for the development of quantum algorithms for data analysis. In this study, we present two innovative data encoding techniques, known as QCrank and QBArt, which exhibit significant quantum parallelism via uniformly controlled rotation gates. The QCrank method encodes a series of real-valued data as rotations on data qubits, resulting in increased storage capacity. On the other hand, QBArt directly incorporates a binary representation of the data within the computational basis, requiring fewer quantum measurements and enabling well-established arithmetic operations on binary data. We showcase various applications of the proposed encoding methods for various data types. Notably, we demonstrate quantum algorithms for tasks such as DNA pattern matching, Hamming weight computation, complex value conjugation, and the retrieval of a binary image with 384 pixels, all executed on the Quantinuum trapped-ion QPU. Furthermore, we employ several cloud-accessible QPUs, including those from IBMQ and IonQ, to conduct supplementary benchmarking experiments.
Here, we present a method for computing transition points of the random cluster model using a generalization of the Newman–Ziff algorithm, a celebrated technique in numerical percolation, to the random cluster model. The new method is straightforward to implement and works for real cluster weight q > 0. Furthermore, results for an arbitrary number of values of q can be found at once within a single simulation. Because the algorithm used to sweep through bond configurations is identical to that of Newman and Ziff, which was conceived for percolation, the method loses accuracy for large lattices when q > 1. However, by sampling the critical polynomial, accurate estimates of critical points in two dimensions can be found using relatively small lattice sizes, which we demonstrate here by computing critical points for non-integer values of q on the square lattice, to compare with the exact solution, and on the unsolved non-planar square matching lattice. The latter results would be much more difficult to obtain using other techniques.
Network Monitoring requires a substantial use of data and error analysis to overcome issues with clusters. Zenoss and Splunk help to monitor system log messages that are reporting issues about the clusters to monitoring services. Infiniband infrastructure on a number of clusters upgraded to ibmon2. ibmon2 requires different filters to report errors to system administrators. Focus for this summer is to: (1) Implement ibmon2 filters on monitoring boxes to report system errors to system administrators using Zenoss and Splunk; (2) Modify and improve scripts for monitoring and administrative usage; (3) Learn more about networks including services and maintenance for high performance computing systems; and (4) Gain a life experience working with professionals under real world situations. Filters were created to account for clusters running ibmon2 v1.0.0-1 10 Filters currently implemented for ibmon2 using Python. Filters look for threshold of port counters. Over certain counts, filters report errors to on-call system administrators and modifies grid to show local host with issue.
Abstract Quantum machine learning (QML) can complement the growing trend of using learned models for a myriad of classification tasks, from image recognition to natural speech processing. There exists the potential for a quantum advantage due to the intractability of quantum operations on a classical computer. Many datasets used in machine learning are crowd sourced or contain some private information, but to the best of our knowledge, no current QML models are equipped with privacy-preserving features. This raises concerns as it is paramount that models do not expose sensitive information. Thus, privacy-preserving algorithms need to be implemented with QML. One solution is to make the machine learning algorithm differentially private, meaning the effect of a single data point on the training dataset is minimized. Differentially private machine learning models have been investigated, but differential privacy has not been thoroughly studied in the context of QML. In this study, we develop a hybrid quantum-classical model that is trained to preserve privacy using differentially private optimization algorithm. This marks the first proof-of-principle demonstration of privacy-preserving QML. The experiments demonstrate that differentially private QML can protect user-sensitive information without signficiantly diminishing model accuracy. Although the quantum model is simulated and tested on a classical computer, it demonstrates potential to be efficiently implemented on near-term quantum devices [noisy intermediate-scale quantum (NISQ)]. The approach’s success is illustrated via the classification of spatially classed two-dimensional datasets and a binary MNIST classification. This implementation of privacy-preserving QML will ensure confidentiality and accurate learning on NISQ technology.
Coupled angular-momentum eigenstates are widely used in atomic and nuclear physics calculations and are building blocks for spin networks and the Schur transform. To combine two angular momenta J 1 and J 2 , forming eigenstates of their total angular momentum J=J 1 +J 2 , we develop a quantum-walk scheme that does not require inputting O(j 3 ) nonzero Clebsch–Gordan (CG) coefficients classically. In fact, our scheme may be regarded as a unitary method for computing CG coefficients on quantum computers with a typical complexity of O(j) and a worst-case complexity of O(j 3 ). Equivalently, our scheme provides decompositions of the dense CG unitary into sparser unitary operations. Our scheme prepares angular-momentum eigenstates using a sequence of Hamiltonians to move an initial state deterministically to desired final states, which are usually highly entangled states in the computational basis. In contrast with usual quantum walks, whose Hamiltonians are prescribed, we engineer the Hamiltonians in su(2)×su(2), which are inspired by, but different from, Hamiltonians that govern magnetic resonances and dipole interactions. To achieve a deterministic preparation of both ket and bra states, we use projection and destructive interference to double pinch the quantum walks, such that each step is a unit-probability population transfer within a two-level system. We test our state preparation scheme on classical computers, reproducing tables of CG coefficients. Finally, we also implement small test problems on current quantum hardware.
The GraphBLAS standard (GraphBlas.org) is being developed to bring the potential of matrix-based graph algorithms to the broadest possible audience. Mathematically, the GraphBLAS defines a core set of matrix-based graph operations that can be used to implement a wide class of graph algorithms in a wide range of programming environments. This study provides an introduction to the mathematics of the GraphBLAS. Graphs represent connections between vertices with edges. Matrices can represent a wide range of graphs using adjacency matrices or incidence matrices. Adjacency matrices are often easier to analyze while incidence matrices are often better for representing data. Fortunately, the two are easily connected by matrix multiplication. A key feature of matrix mathematics is that a very small number of matrix operations can be used to manipulate a very wide range of graphs. This composability of a small number of operations is the foundation of the GraphBLAS. A standard such as the GraphBLAS can only be effective if it has low performance overhead. Finally, performance measurements of prototype GraphBLAS implementations indicate that the overhead is low.
Background: The recent emergence of the H5N1 influenza virus from avian reservoirs has raised concern about future influenza strains of high virulence emerging that could easily infect humans. We analyzed differential gene expression of lung epithelial cells to compare the response to H5N1 infection with a more benign infection with Respiratory Syncytial Virus (RSV). These gene expression data are then used as seeds to find important nodes by using a novel combination of the Gene Ontology database and the Human Network of gene interactions. Additional analysis of the data is conducted by training support vector machines (SVM) with the data and examining the orientations of the optimal hyperplanes generated. Results: Analysis of gene clustering in the Gene Ontology shows no significant clustering of genes unique to H5N1 response at 8 hours post infection. At 24 hours post infection, however, a number of significant gene clusters are found for nodes representing “immune response” and “response to virus” terms. There were no significant clusters of genes in the Gene Ontology for the control (Mock) or RSV experiments that were unique relative to the H5N1 response. The genes found to be most important in distinguishing H5N1 infected cells from the controls using SVM showed a large degree of overlap with the list of significantly regulated genes. However, though none of these genes were members of the GO clusters found to be significant. Conclusions: Characteristics of H5N1 infection compared to RSV infection show several immune response factors that are specific for each of these infections. These include faster timescales within the cell as well as a more focused activation of immunity factors. Many of the genes that are found to be significantly expressed in H5N1 response relative to the control experiments are not found to cluster significantly in the Gene Ontology. These genes are, however, often closely linked to the clustered genes through the Human Network. This may suggest the need for more diverse annotations of these genes and verification of their action in immune response.
Background: The Gene Ontology project supports categorization of gene products according to their location of action, the molecular functions that they carry out, and the processes that they are involved in. Although the ontologies are intentionally developed to be taxon neutral, and to cover all species, there are inherent taxon specificities in some branches. For example, the process ‘lactation’ is specific to mammals and the location ‘mitochondrion’ is specific to eukaryotes. The lack of an explicit formalization of these constraints can lead to errors and inconsistencies in automated and manual annotation. Results: We have formalized the taxonomic constraints implicit in some GO classes, and specified these at various levels in the ontology. We have also developed an inference system that can be used to check for violations of these constraints in annotations. Using the constraints in conjunction with the inference system, we have detected and removed errors in annotations and improved the structure of the ontology. Conclusions: Detection of inconsistencies in taxon-specificity enables gradual improvement of the ontologies, the annotations, and the formalized constraints. This is progressively improving the quality of our data. The full system is available for download, and new constraints or proposed changes to constraints can be submitted online at https://sourceforge.net/tracker/?atid=605890&group_id=36855.
Background: We developed an extendable open-source Loop-mediated isothermal AMPlification (LAMP) signature design program called LAVA (LAMP Assay Versatile Analysis). LAVA was created in response to limitations of existing LAMP signature programs. Results: LAVA identifies combinations of six primer regions for basic LAMP signatures, or combinations of eight primer regions for LAMP signatures with loop primers, which can be used as LAMP signatures. The identified primers are conserved among target organism sequences. Primer combinations are optimized based on lengths, melting temperatures, and spacing among primer sites. We compare LAMP signature candidates for Staphylococcus aureus created both by LAVA and by PrimerExplorer. We also include signatures from a sample run targeting all strains of Mycobacterium tuberculosis. Conclusions: We have designed and demonstrated new software for identifying signature candidates appropriate for LAMP assays. The software is available for download at http://lava-dna.googlecode.com/.