Search NASA⌕ Search

SEARCH · Search NASA

Results for “kernel method”

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.

At least 1,225 records · Page 68

Efficient Generalizable Deep Learning

Stochastic optimization is a fundamental field of research for machine learning. Stochastic gradient descent (SGD) and related methods provide a feasible means to train complicated prediction models over large datasets. SGD, however, does not explicitly address the problem of overfitting, which can lead to predictions that perform poorly on new data. This difference between loss performance on unseen testing data verses that of training data defines the generalization gap of a model. We introduce a new computational kernel called Stochastic Hessian Projection (SHP) that uses a maximum likelihood framework to simultaneously estimate gradient noise covariance and local curvature of the loss function. Our analysis illustrates that these quantities affect the evolution of parameter uncertainty and therefore generalizability. We show how these computations allow us to predict the generalization gap without requiring holdout data. Explicitly assessing this metric for generalizability during training may improve machine learning predictions when data is scarce and understanding prediction variability is critical.

42 ENGINEERING↗

EPMA-Based Mass Balance Method for Quantitative Fission Product Distribution Comparison between TRISO Particles

Two irradiated AGR-2 TRISO particles were chosen to demonstrate a recently developed mass balance technique in which EPMA-generated concentration data was used to determine fission product mass on a layer-by-layer basis in TRISO particles. EPMA-calculated fission product masses for most fission products in the two particles were within +/- 20% of their ORIGEN-modelled masses. Results show that Sr, Ba, and Eu accumulate preferentially in the carbon-rich kernel periphery on the particles’ side that lacks a gap between the buffer and IPyC. In addition, the more mobile elements--Cs, Sr, and Pd, accumulate in greater quantity in the outer layers of particle AGR2-223-RS34 compared to particle AGR2-223-RS06, which has relatively more of those elements’ mass located in the kernel and kernel periphery, suggesting enhanced fission product transport in particle AGR2-223-RS34. This model can be used better understand and test fission product transport in TRISO particles.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

Molecular-orbital-based machine learning for open-shell and multi-reference systems with kernel addition Gaussian process regression

We introduce a novel machine learning strategy, kernel addition Gaussian process regression (KA-GPR), in molecular-orbital-based machine learning (MOB-ML) to learn the total correlation energies of general electronic structure theories for closed- and open-shell systems by introducing a machine learning strategy. The learning efficiency of MOB-ML(KA-GPR) is the same as the original MOB-ML method for the smallest criegee molecule, which is a closed-shell molecule with multi-reference characters. In addition, the prediction accuracies of different small free radicals could reach the chemical accuracy of 1 kcal/mol by training on one example structure. Accurate potential energy surfaces for the H10 chain (closed-shell) and water OH bond dissociation (open-shell) could also be generated by MOB-ML(KA-GPR). To explore the breadth of chemical systems that KA-GPR can describe, we further apply MOB-ML to accurately predict the large benchmark datasets for closed- (QM9, QM7b-T, and GDB-13-T) and open-shell (QMSpin) molecules.

Chemistry↗

A Framework for Error-Bounded Approximate Computing, with an Application to Dot Products

Approximate computing techniques, which trade off the computation accuracy of an algorithm for better performance and energy efficiency, have been successful in reducing computation and power costs in several domains. However, error sensitive applications in high-performance computing are unable to benefit from existing approximate computing strategies that are not developed with guaranteed error bounds. While approximate computing techniques can be developed for individual high-performance computing applications by domain specialists, this often requires additional theoretical analysis and potentially extensive software modification. Hence, the development of low-level error-bounded approximate computing strategies that can be introduced into any high-performance computing application without requiring additional analysis or significant software alterations is desirable. In this paper, we provide a contribution in this direction by proposing a general framework for designing error-bounded approximate computing strategies and apply it to the dot product kernel to develop \bf qdot---an error-bounded approximate dot product kernel. Following the introduction of qdot, here we perform a theoretical analysis that yields a deterministic bound on the relative approximation error introduced by qdot. Empirical tests are performed to illustrate the tightness of the derived error bound and to demonstrate the effectiveness of qdot on a synthetic dataset, as well as two scientific benchmarks---the conjugate gradient (CG) and power methods. In some instances, using qdot for the dot products in CG can result in many components being quantized to half precision without increasing the iteration count required for convergence to the same solution as CG using a double precision dot product.

97 MATHEMATICS AND COMPUTING↗

Approximating linearizations for nonlinear systems

The following problem is examined: given a nonlinear control system dot-x(t) = f(x(t)) + the sum to m terms(i=1) u sub i (t)g sub i (x(t)) on R(n) and a point x(0) in R(n), approximate the system near x(0) by a linear system. One approach is to use the usual Taylor series linearization. However, the controllability properties of both the nonlinear and linear systems depend on certain Lie brackets of the vector field under consideration. This suggests that a linear approximation based on Lie bracket matching should be constructed at x(0). In general, the linearizations based on the Taylor method and the Lie bracket approach are different. However, under certain mild assumptions, it is shown that there is a coordinate system for R(n) near x(0) in which these two types of linearizations agree. The importance of this agreement is indicated by examining the time responses of the nonlinear system and its linear approximation and comparing the lower order kernels in Volterra expansions of each.

Hunt, L. R.↗

Approximating linearizations for nonlinear systems

The following problem is examined: given a nonlinear control system dot-x(t) = f(x/t/) + the sum to m terms (i = 1) u sub i (t)g sub i (x/t/) on R(n) and a point x(0) in R(n), approximate the system near x(0) by a linear system. One approach is to use the usual Taylor series linearization. However, the controllability properties of both the nonlinear and linear systems depend on certain Lie brackets of the vector field under consideration. This suggests that a linear approximation based on Lie bracket matching should be constructed at x(0). In general, the linearizations based on the Taylor method and the Lie bracket approach are different. However, under certain mild assumptions, it is shown that there is a coordinate system for R(n) near x(0) in which these two types of linearizations agree. The importance of this agreement is indicated by examining the time responses of the nonlinear system and its linear approximation and comparing the lower order kernels in Volterra expansions of each.

Hunt, L. R.↗

Optimization of Support Vector Machine (SVM) for Object Classification

The Support Vector Machine (SVM) is a powerful algorithm, useful in classifying data into species. The SVMs implemented in this research were used as classifiers for the final stage in a Multistage Automatic Target Recognition (ATR) system. A single kernel SVM known as SVMlight, and a modified version known as a SVM with K-Means Clustering were used. These SVM algorithms were tested as classifiers under varying conditions. Image noise levels varied, and the orientation of the targets changed. The classifiers were then optimized to demonstrate their maximum potential as classifiers. Results demonstrate the reliability of SVM as a method for classification. From trial to trial, SVM produces consistent results.

support vector machice (SVM)↗

MO-FG-BRA-01: Development of An Image-Guided Dosimetric Planning System for Injectable Brachytherapy Using ELP Nanoparticles

Purpose: To develop, validate, and evaluate a methodology for determining dosimetry for intratumoral injections of elastin-like-polypeptide (ELP) brachytherapy nanoparticles. These organic-polymer-based nanoparticles are injectable, biodegradable, and genetically tunable. We present a genetically encoded polymer-solution, composed of novel radiolabeled-ELP nanoparticles that are custom-designed to self-assemble into a local source upon intratumoral injection. Our preliminary results of a small animal study demonstrate 100% tumor response, effective radionuclide retention-rates, strong in vivo stability, and no polymer-induced toxicities. While our approach is therefore highly promising for improved brachytherapy, the current workflow lacks a dosimetry framework. Methods: We are developing a robust software framework that provides image-guided dosimetric-planning capabilities for ELP brachytherapy. The user graphically places ELP injection sites within a µCT-planning-image, and independently defines each injection volume, concentration, and radioisotope to be used. The resulting internal dosimetry is then pre-determined by first modeling post-injection ELP advection-diffusion, and then calculating the resulting dose distribution based on a point- dose-kernel-convolution algorithm. We have experimentally measured ELP steady-state concentrations via µSPECT acquisition, and validated our dose calculation algorithm against Monte Carlo simulations of several radioactivity distributions. Finally, we have investigated potential advantages and limitations of various ELP injection parameters. Results: The µSPECT results demonstrated inhomogeneous steady-state distributions of ELP in tissue, and Monte Carlo radioactivity distributions were designed accordingly. Our algorithm yielded a root-mean-square-error of less than 2% for each distribution tested (average root-mean-square-error was 0.73%). Dose-Volume-Histogram analysis of five different plans showed how strategic injection placement, and an injection volume-tapering technique, could be used to achieve D95% target coverage. Conclusion: We have preliminarily developed a novel planning framework for ELP brachytherapy. Its dosimetry accuracy has been validated against Monte Carlo, and we have started to investigate the potential advantages of injection-based planning. This system, once fully developed, will serve as the technical foundation for our novel approach.

60 APPLIED LIFE SCIENCES↗

ASC Weekly News Notes

Unified collision operator demonstrated for both radiation transport and PIC-DSMC. A side-by-side comparison between the DSMC method and the radiation transport method was conducted for photon attenuation in the atmosphere over 2 kilometers in physical distance with a reduction of photon density of six orders of magnitude. Both DSMC and traditional radiation transport agreed with theory to two digits. This indicates that PIC-DSMC operators can be unified with the radiation transport collision operators into a single code base and that physics kernels can remain unique to the actual collision pairs. This simulation example provides an initial validation of the unified collision theory approach that will later be implemented into EMPIRE.

54 ENVIRONMENTAL SCIENCES↗

Enhanced Approximate Nearest Neighbor via Local Area Focused Search.

Approximate Nearest Neighbor (ANN) algorithms are increasingly important in machine learning, data mining, and image processing applications. There is a large family of space- partitioning ANN algorithms, such as randomized KD-Trees, that work well in practice but are limited by an exponential increase in similarity comparisons required to optimize recall. Additionally, they only support a small set of similarity metrics. We present Local Area Fo- cused Search (LAFS), a method that enhances the way queries are performed using an existing ANN index. Instead of a single query, LAFS performs a number of smaller (fewer similarity comparisons) queries and focuses on a local neighborhood which is refined as candidates are identified. We show that our technique improves performance on several well known datasets and is easily extended to general similarity metrics using kernel projection techniques.

97 MATHEMATICS AND COMPUTING↗

An Algorithm to Generate Deep-Layer Temperatures from Microwave Satellite Observations for the Purpose of Monitoring Climate Change

An algorithm for generating deep-layer mean temperatures from satellite-observed microwave observations is presented. Unlike traditional temperature retrieval methods, this algorithm does not require a first guess temperature of the ambient atmosphere. By eliminating the first guess a potentially systematic source of error has been removed. The algorithm is expected to yield long-term records that are suitable for detecting small changes in climate. The atmospheric contribution to the deep-layer mean temperature is given by the averaging kernel. The algorithm computes the coefficients that will best approximate a desired averaging kernel from a linear combination of the satellite radiometer's weighting functions. The coefficients are then applied to the measurements to yield the deep-layer mean temperature. Three constraints were used in deriving the algorithm: (1) the sum of the coefficients must be one, (2) the noise of the product is minimized, and (3) the shape of the approximated averaging kernel is well-behaved. Note that a trade-off between constraints 2 and 3 is unavoidable. The algorithm can also be used to combine measurements from a future sensor (i.e., the 20-channel Advanced Microwave Sounding Unit (AMSU)) to yield the same averaging kernel as that based on an earlier sensor (i.e., the 4-channel Microwave Sounding Unit (MSU)). This will allow a time series of deep-layer mean temperatures based on MSU measurements to be continued with AMSU measurements. The AMSU is expected to replace the MSU in 1996.

Goldberg, Mitchell D.↗

Using SpF to Achieve Petascale for Legacy Pseudospectral Applications

Pseudospectral (PS) methods possess a number of characteristics (e.g., efficiency, accuracy, natural boundary conditions) that are extremely desirable for dynamo models. Unfortunately, dynamo models based upon PS methods face a number of daunting challenges, which include exposing additional parallelism, leveraging hardware accelerators, exploiting hybrid parallelism, and improving the scalability of global memory transposes. Although these issues are a concern for most models, solutions for PS methods tend to require far more pervasive changes to underlying data and control structures. Further, improvements in performance in one model are difficult to transfer to other models, resulting in significant duplication of effort across the research community. We have developed an extensible software framework for pseudospectral methods called SpF that is intended to enable extreme scalability and optimal performance. Highlevel abstractions provided by SpF unburden applications of the responsibility of managing domain decomposition and load balance while reducing the changes in code required to adapt to new computing architectures. The key design concept in SpF is that each phase of the numerical calculation is partitioned into disjoint numerical kernels that can be performed entirely inprocessor. The granularity of domain decomposition provided by SpF is only constrained by the datalocality requirements of these kernels. SpF builds on top of optimized vendor libraries for common numerical operations such as transforms, matrix solvers, etc., but can also be configured to use open source alternatives for portability. SpF includes several alternative schemes for global data redistribution and is expected to serve as an ideal testbed for further research into optimal approaches for different network architectures. In this presentation, we will describe our experience in porting legacy pseudospectral models, MoSST and DYNAMO, to use SpF as well as present preliminary performance results provided by the improved scalability.

DYNAMO↗

Subgrid modelling for two-dimensional turbulence using neural networks

Here in this investigation, a data-driven turbulence closure framework is introduced and deployed for the subgrid modelling of Kraichnan turbulence. The novelty of the proposed method lies in the fact that snapshots from high-fidelity numerical data are used to inform artificial neural networks for predicting the turbulence source term through localized grid-resolved information. In particular, our proposed methodology successfully establishes a map between inputs given by stencils of the vorticity and the streamfunction along with information from two well-known eddy-viscosity kernels. Through this we predict the subgrid vorticity forcing in a temporally and spatially dynamic fashion. Our study is both a priori and a posteriori in nature. In the former, we present an extensive hyper-parameter optimization analysis in addition to learning quantification through probability-density-function-based validation of subgrid predictions. In the latter, we analyse the performance of our framework for flow evolution in a classical decaying two-dimensional turbulence test case in the presence of errors related to temporal and spatial discretization. Statistical assessments in the form of angle-averaged kinetic energy spectra demonstrate the promise of the proposed methodology for subgrid quantity inference. In addition, it is also observed that some measure of a posteriori error must be considered during optimal model selection for greater accuracy. The results in this article thus represent a promising development in the formalization of a framework for generation of heuristic-free turbulence closures from data.

42 ENGINEERING↗

Extended Lagrangian Born–Oppenheimer molecular dynamics using a Krylov subspace approximation

It is shown how the electronic equations of motion in extended Lagrangian Born–Oppenheimer molecular dynamics simulations can be integrated using low-rank approximations of the inverse Jacobian kernel. This kernel determines the metric tensor in the harmonic oscillator extension of the Lagrangian that drives the evolution of the electronic degrees of freedom. The proposed kernel approximation is derived from a pseudoinverse of a low-rank estimate of the Jacobian, which is expressed in terms of a generalized set of directional derivatives with directions that are given from a Krylov subspace approximation. The approach allows a tunable and adaptive approximation that can take advantage of efficient preconditioning techniques. The proposed kernel approximation for the integration of the electronic equations of motion makes it possible to apply extended Lagrangian first-principles molecular dynamics simulations to a broader range of problems, including reactive chemical systems with numerically sensitive and unsteady charge solutions. This can be achieved without requiring exact full calculations of the inverse Jacobian kernel in each time step or relying on iterative non-linear self-consistent field optimization of the electronic ground state prior to the force evaluations as in regular direct Born–Oppenheimer molecular dynamics. We note the low-rank approximation of the Jacobian is directly related to Broyden’s class of quasi-Newton algorithms and Jacobian-free Newton–Krylov methods and provides a complementary formulation for the solution of nonlinear systems of equations.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Flow Analysis Tool White Paper

Faster networks are continually being built to accommodate larger data transfers. While it is intuitive to think that implementing faster networks will result in higher throughput rates, this is often not the case. There are many elements involved in data transfer, many of which are beyond the scope of the network itself. Although networks may get bigger and support faster technologies, the presence of other legacy components, such as older application software or kernel parameters, can often cause bottlenecks. Engineers must be able to identify when data flows are reaching a bottleneck that is not imposed by the network and then troubleshoot it using the tools available to them. The current best practice is to collect as much information as possible on the network traffic flows so that analysis is quick and easy. Unfortunately, no single method of collecting this information can sufficiently capture the whole endto- end picture. This becomes even more of a hurdle when large, multi-user systems are involved. In order to capture all the necessary information, multiple data sources are required. This paper presents a method for developing a flow analysis tool to effectively collect network flow data from multiple sources and provide that information to engineers in a clear, concise way for analysis. The purpose of this method is to collect enough information to quickly (and automatically) identify poorly performing flows along with the cause of the problem. The method involves the development of a set of database tables that can be populated with flow data from multiple sources, along with an easyto- use, web-based front-end interface to help network engineers access, organize, analyze, and manage all the information.

Boscia, Nichole K.↗

Testing of the Support Vector Machine for Binary-Class Classification

The Support Vector Machine is a powerful algorithm, useful in classifying data in to species. The Support Vector Machines implemented in this research were used as classifiers for the final stage in a Multistage Autonomous Target Recognition system. A single kernel SVM known as SVMlight, and a modified version known as a Support Vector Machine with K-Means Clustering were used. These SVM algorithms were tested as classifiers under varying conditions. Image noise levels varied, and the orientation of the targets changed. The classifiers were then optimized to demonstrate their maximum potential as classifiers. Results demonstrate the reliability of SMV as a method for classification. From trial to trial, SVM produces consistent results

autonomous target recognition systemr↗

Model-based versus specific dosimetry in diagnostic context: Comparison of three dosimetric approaches

Purpose: The dosimetric assessment of novel radiotracers represents a legal requirement in most countries. While the techniques for the computation of internal absorbed dose in a therapeutic context have made huge progresses in recent years, in a diagnostic scenario the absorbed dose is usually extracted from model-based lookup tables, most often derived from International Commission on Radiological Protection (ICRP) or Medical Internal Radiation Dose (MIRD) Committee models. The level of approximation introduced by these models may impact the resulting dosimetry. The aim of this work is to establish whether a more refined approach to dosimetry can be implemented in nuclear medicine diagnostics, by analyzing a specific case. Methods: The authors calculated absorbed doses to various organs in six healthy volunteers administered with flutemetamol ({sup 18}F) injection. Each patient underwent from 8 to 10 whole body 3D PET/CT scans. This dataset was analyzed using a Monte Carlo (MC) application developed in-house using the toolkit GATE that is capable to take into account patient-specific anatomy and radiotracer distribution at the voxel level. They compared the absorbed doses obtained with GATE to those calculated with two commercially available software: OLINDA/EXM and STRATOS implementing a dose voxel kernel convolution approach. Results: Absorbed doses calculated with GATE were higher than those calculated with OLINDA. The average ratio between GATE absorbed doses and OLINDA’s was 1.38 ± 0.34 σ (from 0.93 to 2.23). The discrepancy was particularly high for the thyroid, with an average GATE/OLINDA ratio of 1.97 ± 0.83 σ for the six patients. Differences between STRATOS and GATE were found to be higher. The average ratio between GATE and STRATOS absorbed doses was 2.51 ± 1.21 σ (from 1.09 to 6.06). Conclusions: This study demonstrates how the choice of the absorbed dose calculation algorithm may introduce a bias when gamma radiations are of importance, as is the case in nuclear medicine diagnostics.

60 APPLIED LIFE SCIENCES↗