Search NASA⌕ Search

SEARCH · Search NASA

Results for “Distributed training”

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 253 records · Page 14

Data-centric machine learning in quantum information science

Abstract We propose a series of data-centric heuristics for improving the performance of machine learning systems when applied to problems in quantum information science. In particular, we consider how systematic engineering of training sets can significantly enhance the accuracy of pre-trained neural networks used for quantum state reconstruction without altering the underlying architecture. We find that it is not always optimal to engineer training sets to exactly match the expected distribution of a target scenario, and instead, performance can be further improved by biasing the training set to be slightly more mixed than the target. This is due to the heterogeneity in the number of free variables required to describe states of different purity, and as a result, overall accuracy of the network improves when training sets of a fixed size focus on states with the least constrained free variables. For further clarity, we also include a ‘toy model’ demonstration of how spurious correlations can inadvertently enter synthetic data sets used for training, how the performance of systems trained with these correlations can degrade dramatically, and how the inclusion of even relatively few counterexamples can effectively remedy such problems.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Accurate and rapid acoustic damage characterization in complex structures using sparse sensor networks and deep learning models

Damage diagnosis in critical components is essential for ensuring the safety and reliability of operations across industries, spanning manufacturing, aerospace, and energy. Traditional acoustic nondestructive testing methods primarily focus on detecting defects through the direct scattering of single-mode incident waves from the damage, which limit their applicability to simple structures and small inspection areas. Our earlier research demonstrated that machine learning algorithms combined with sparse sensor networks can identify critical defect signatures even from multiply scattered, multi-mode acoustic signals, indicating the potential for improved defect inspection in complex, real-world structures. In this work, we demonstrate the successful implementation of this approach in a fixed sensor configuration to rapidly and accurately detect simulated defects in a geometrically complex, real-world structure, a brake rotor hub. Three different types of defects were physically simulated on the surface of the hub, and the collected data were used to train an autoencoder-based deep learning model. Two models were tested, one using single measurements and the other using multiple measurements taking advantage of the spatial distribution of the sensor network. After training, the multi-measurement model achieved 100 % accuracy in identifying, classifying, and locating unseen, unique damages. This work illustrates the potential of the proposed method for a wide range of industrial applications.

36 MATERIALS SCIENCE↗

Neural network approximation of Bayesian models for the inference of ion and electron temperature profiles at W7-X

Here, we describe a method for training a neural network (NN) to approximate the full model Bayesian inference of plasma profiles from x-ray imaging diagnostic measurements. The modeling is carried out within the Minerva Bayesian modeling framework where models are defined as a set of assumptions, prior beliefs on parameter values and physics knowledge. The goal is to use NNs for fast ion and electron temperature profile inversion from measured image data. The NN is trained solely on artificial data generated by sampling from the joint distribution of the free parameters and model predictions. The training is carried out in such a way that the mapping learned by the network constitutes an approximation of the full model Bayesian inference. The analysis is carried out on images constituted of 20 × 195 pixels corresponding to binned lines of sight and spectral channels, respectively. Through the full model inference, it is possible to infer electron and ion temperature profiles as well as impurity density profiles. When the network is utilized for the inference of the temperature profiles, the analysis time can be reduced down to a few tens of microseconds for a single time point, which is a drastic improvement if compared to the ≈4 h long Bayesian inference. The procedure developed for the generation of the training set does not rely on diagnostic-specific features, and therefore it is in principle applicable to any other model developed within the Minerva framework. The trained NN has been tested on data collected during the first operational campaign at W7-X, and compared to the full model Bayesian inference results.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Predicting Mechanical Properties from Microstructure Images in Fiber-Reinforced Polymers Using Convolutional Neural Networks

Evaluating the mechanical response of fiber-reinforced composites can be extremely time-consuming and expensive. Machine learning (ML) techniques offer a means for faster predictions via models trained on existing input–output pairs and have exhibited success in composite research. This paper explores a fully convolutional neural network modified from StressNet, which was originally used for linear elastic materials, and extended here for a non-linear finite element (FE) simulation to predict the stress field in 2D slices of segmented tomography images of a fiber-reinforced polymer specimen. The network was trained and evaluated on data generated from the FE simulations of the exact microstructure. The testing results show that the trained network accurately captures the characteristics of the stress distribution, especially on fibers, solely from the segmented microstructure images. The trained model can make predictions within seconds in a single forward pass on an ordinary laptop, given the input microstructure, compared to 92.5 h to run the full FE simulation on a high-performance computing cluster. These results show promise in using ML techniques to conduct fast structural analysis for fiber-reinforced composites and suggest a corollary that the trained model can be used to identify the location of potential damage sites in fiber-reinforced polymers.

Sun, Yixuan (ORCID:0000000311093380)↗

Using convolutional neural networks to accelerate three-dimensional coherent synchrotron radiation computations

Calculating the effects of coherent synchrotron radiation (CSR) is one of the most computationally expensive tasks in accelerator physics. Here, we use convolutional neural networks (CNNs), along with a latent conditional diffusion (LCD) model, trained on physics-based simulations to speed up calculations. Specifically, we produce the 3D CSR wakefields generated by electron bunches in circular orbit in the steady-state condition. Two datasets are used for training and testing the models: wakefields generated by three-dimensional Gaussian electron distributions and wakefields from a sum of up to 25 three-dimensional Gaussian distributions. The CNNs are able to accurately produce the 3D wakefields ∼250–1000 times faster than the numerical calculations, while the LCD achieves a gain of a factor of ∼34. We also test the extrapolation and out-of-distribution generalization ability of the models. They generalize well on distributions with larger spreads than what they were trained on but struggle with smaller spreads.

43 PARTICLE ACCELERATORS↗

An Observational and modeling strategy to investigate the impact of remote sources on local air quality: A Houston, Texas case study from the Second Texas Air Quality Study (TEXAQS II)

Quantifying the impacts of remote sources on individual air quality exceedances remains a significant challenge for air quality forecasting. One goal of the 2006 Texas Air Quality Study (TEXAQS II) was to assess the impact of distant sources on air quality in east Texas. From 23-30 August 2006, retrievals of tropospheric carbon monoxide (CO) from NASA’s Atmospheric InfraRed Sounder (AIRS) reveal the transport of CO from fires in the United States Pacific Northwest to Houston, Texas. This transport occurred behind a cold front and contributed to the worst ozone exceedance period of the summer in the Houston area. We present supporting satellite observations from the NASA A-Train constellation of the vertical distribution of smoke aerosols and CO. Ground-based in situ CO measurements in Oklahoma and Texas track the CO plume as it moves south and indicate mixing of the aloft plume to the surface by turbulence in the nocturnal boundary layer and convection during the day. Ground-based aerosol speciation and lidar observations do not find appreciable smoke aerosol transport for this case. However, MODIS aerosol optical depths and model simulations indicate some smoke aerosols were transported from the Pacific Northwest through Texas to the Gulf of Mexico. Chemical transport and forward trajectory models confirm the three major observations: (1) the AIRS envisioned CO transport, (2) the satellite determined smoke plume height, and (3) the timing of the observed surface CO increases. Further, the forward trajectory simulations find two of the largest Pacific Northwest fires likely had the most significant impact.

54 ENVIRONMENTAL SCIENCES↗

Multiscale Flow for robust and optimal cosmological analysis

We propose Multiscale Flow, a generative Normalizing Flow that creates samples and models the field-level likelihood of two-dimensional cosmological data such as weak lensing. Multiscale Flow uses hierarchical decomposition of cosmological fields via a wavelet basis and then models different wavelet components separately as Normalizing Flows. The log-likelihood of the original cosmological field can be recovered by summing over the log-likelihood of each wavelet term. This decomposition allows us to separate the information from different scales and identify distribution shifts in the data such as unknown scale-dependent systematics. The resulting likelihood analysis can not only identify these types of systematics, but can also be made optimal, in the sense that the Multiscale Flow can learn the full likelihood at the field without any dimensionality reduction. We apply Multiscale Flow to weak lensing mock datasets for cosmological inference and show that it significantly outperforms traditional summary statistics such as power spectrum and peak counts, as well as machine learning–based summary statistics such as scattering transform and convolutional neural networks. We further show that Multiscale Flow is able to identify distribution shifts not in the training data such as baryonic effects. Finally, we demonstrate that Multiscale Flow can be used to generate realistic samples of weak lensing data.

79 ASTRONOMY AND ASTROPHYSICS↗

Assessing Machine Learning as a Tool to Explain Variance in Deployed Photovoltaic (PV) System Degradation

Degradation remains a large uncertainty in forecasting production for PV plants, creating significant risk for developers and financiers. This study aims to quantify the distribution and drivers of degradation across 10,000 PV systems deployed for distributed or utility generation by training a machine learning model to predict year-over-year degradation rates from metadata characteristics. A combination of K-Means clustering and random forest regressor were found to associate multiple metadata features as potential drivers of degradation, including module characteristics, system design, and climate features. From this, it is inferred that if machine learning is able to find complex patterns between metadata features and system performance loss, such methods can be employed to help developers and financiers make data-informed decisions when estimating long-term energy production forecasts in financial models.

Dunn, Jimmy C.↗

Mass agnostic jet taggers

Searching for new physics in large data sets needs a balance between two competing effects—signal identification vs background distortion. In this work, we perform a systematic study of both single variable and multivariate jet tagging methods that aim for this balance. The methods preserve the shape of the background distribution by either augmenting the training procedure or the data itself. Multiple quantitative metrics to compare the methods are considered, for tagging 2-, 3-, or 4-prong jets from the QCD background. This is the first study to show that the data augmentation techniques of Planing and PCA based scaling deliver similar performance as the augmented training techniques of Adversarial NN and uBoost, but are both easier to implement and computationally cheaper.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

A Business Case for Home Performance Contracting

This report was prepared by PNNL for the DOE Building America program. The report provides information for businesses considering entering the home performance contracting industry. Metrics discussed include industry trends and drivers, specific points of entry, business models, startup costs, and marketing strategies. The report includes detailed analysis of eight businesses around the country that have successfully entered the home performance contracting industry. Data is provided on their financial structures, program participation, marketing efforts, and staff training. This report will be distributed via the DOE Building America website, www.buildingamerica.gov. Individual case studies will also be cleared separately.

Building America↗

Staghorn: An Automated Large-Scale Distributed System Analysis Platform

Conducting experiments on large-scale distributed computing systems is becoming significantly easier with the assistance of emulation. Researchers can now create a model of a distributed computing environment and then generate a virtual, laboratory copy of the entire system composed of potentially thousands of virtual machines, switches, and software. The use of real software, running at clock rate in full virtual machines, allows experiments to produce meaningful results without necessitating a full understanding of all model components. However, the ability to inspect and modify elements within these models is bound by the limitation that such modifications must compete with the model, either running in or alongside it. This inhibits entire classes of analyses from being conducted upon these models. We developed a mechanism to snapshot an entire emulation-based model as it is running. This allows us to \freeze time" and subsequently fork execution, replay execution, modify arbitrary parts of the model, or deeply explore the model. This snapshot includes capturing packets in transit and other input/output state along with the running virtual machines. We were able to build this system in Linux using Open vSwitch and Kernel Virtual Machines on top of Sandia's emulation platform Firewheel. This primitive opens the door to numerous subsequent analyses on models, including state space exploration, debugging distributed systems, performance optimizations, improved training environments, and improved experiment repeatability.

97 MATHEMATICS AND COMPUTING↗

Landsat-D thematic mapper simulation using aircraft multispectral scanner data

A simulation of imagery from the upcoming Landsat-D Thematic Mapper was accomplished by using selected channels of aircraft 24-channel multispectral scanner data. The purpose was to simulate Thematic Mapper 30-meter resolution imagery, to compare its spectral quality with the original aircraft MSS data, and to determine changes in thematic classification accuracy for the simulated imagery. The original resolution of approximately 7.5 meters IFOV and simulated resolution of 15, 30, and 60 meters were used to indicate the trend of spectral quality and classification accuracy. The study was based in a 6.5 square kilometer area of urban Los Angeles having a diversity of land use. The original imagery was reduced in resolution by two related methods: pixel matrix averaging, and matrix smoothing with a unity box filter, followed by matrix averaging. Thematic land use classification using training sites and a Bayesian maximum-likelihood algorithm was performed at three levels of standard deviation - 1.0, 2.0, and 3.0 sigma. Plots of relative standard deviation showed that for larger training sites with a normal distribution of data, as the resolution decreased, the distribution range of density values decreased. Also, the classification accuracies for three levels of standard deviation increased as resolution decreased. However, the indication is that a point of diminishing returns had been reached, and 30 meters IFOV should be the best for multispectral classification of urban scenes.

Clark, J.↗

Manual for the Jet Event and Background Simulation Library(JEBSimLib)

Jets are the collimated streams of particles resulting from hard scattering in the initial state of high-energy collisions. In heavy-ion collisions, jets interact with the quark-gluon plasma (QGP) before freezeout, providing a probe into the internal structure and properties of the QGP. In order to study jets, background must be subtracted from the measured event, potentially introducing a bias. We aim to understand and quantify this subtraction bias. PYTHIA, a library to simulate pure jet events, is used to simulate a model for a signature with one pure jet (a photon) and one quenched jet, where all quenched particle momenta are reduced by a user-de ned constant fraction. Background for the event is simulated using multiplicity values generated by the TRENTO initial state model of heavy-ion collisions fed into a thermal model consisting of a 3-dimensional Boltzmann distribution for particle types and momenta. Data from the simulated events is used to train a statistical model, which computes a posterior distribution of the quench factor for a data set. The model was tested rst on pure jet events and then on full events including the background. This model will allow for a quantitative determination of biases induced by various methods of background subtraction.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Manual for the Jet Event and Background Simulation Library

Jets are the collimated streams of particles resulting from hard scattering in the initial state of high-energy collisions. In heavy-ion collisions, jets interact with the quark-gluon plasma (QGP) before freezeout, providing a probe into the internal structure and properties of the QGP. In order to study jets, background must be subtracted from the measured event, potentially introducing a bias. We aim to understand and quantify this subtraction bias. PYTHIA, a library to simulate pure jet events, is used to simulate a model for a signature with one pure jet (a photon) and one quenched jet, where all quenched particle momenta are reduced by a user-de ned constant fraction. Background for the event is simulated using multiplicity values generated by the TRENTO initial state model of heavy-ion collisions fed into a thermal model consisting of a 3-dimensional Boltzmann distribution for particle types and momenta. Data from the simulated events is used to train a statistical model, which computes a posterior distribution of the quench factor for a data set. The model was tested rst on pure jet events and then on full events including the background. This model will allow for a quantitative determination of biases induced by various methods of background subtraction.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Balanced Expertise Distribution in Remote Ultrasound Imaging Aboard The International Space Station (ISS)

Astronaut training for ISS operations usually ensures independent performance. With small crew size same crews also conduct all science work onboard. With diverse backgrounds, a good "match" between the existing and required skills can only be anecdotal. Furthermore, full proficiency in most of the complex tasks can be attained only through long training and practice, which may not be justified and may be impossible given the scarcity of training time. To enable a number of operational and science advancements, authors have developed a new approach to expertise distribution in time and among the space and ground personnel. Methods: As part of NASA Operational Ultrasound Project (1998-2003) and the NASA-solicited experiment "Advanced Diagnostic Ultrasound in Microgravity-ADUM" (P.I. -S.D., ongoing), the authors have created a "Balanced Expertise Distribution" approach to perform complex ultrasound imaging tasks on ISS for both operational and science use. The four components of expertise are a) any pre-existing pertinent expertise; b) limited preflight training c) adaptive onboard proficiency enhancement tools; d) real-time ' guidance from the ground. Throughout the pre-flight training and flight time preceding the experiments, the four components are shaped in a dynamic fashion to meet in an optimum combination during the experiment sessions. Results: Procedure validation sessions and feasibility studies have given encouraging results. While several successful real-time remote guidance sessions have been conducted on ISS, Expedition 8 is the first to use an "on-orbit proficiency enhancement" tool. Conclusions: In spite of severely limited training time, daring peer-reviewed research and operational enhancements are feasible through a balanced distribution of expertise in time, as well as among the crewmembers and ground personnel. This approach shows great promise for biomedical research, but may be applicable for other areas of micro gravity-based science

Sargsyan, Ashot↗

TH-CD-202-07: A Methodology for Generating Numerical Phantoms for Radiation Therapy Using Geometric Attribute Distribution Models

Purpose: A persistent challenge for the quality assessment of radiation therapy treatments (e.g. contouring accuracy) is the absence of the known, ground truth for patient data. Moreover, assessment results are often patient-dependent. Computer simulation studies utilizing numerical phantoms can be performed for quality assessment with a known ground truth. However, previously reported numerical phantoms do not include the statistical properties of inter-patient variations, as their models are based on only one patient. In addition, these models do not incorporate tumor data. In this study, a methodology was developed for generating numerical phantoms which encapsulate the statistical variations of patients within radiation therapy, including tumors. Methods: Based on previous work in contouring assessment, geometric attribute distribution (GAD) models were employed to model both the deterministic and stochastic properties of individual organs via principle component analysis. Using pre-existing radiation therapy contour data, the GAD models are trained to model the shape and centroid distributions of each organ. Then, organs with different shapes and positions can be generated by assigning statistically sound weights to the GAD model parameters. Organ contour data from 20 retrospective prostate patient cases were manually extracted and utilized to train the GAD models. As a demonstration, computer-simulated CT images of generated numerical phantoms were calculated and assessed subjectively and objectively for realism. Results: A cohort of numerical phantoms of the male human pelvis was generated. CT images were deemed realistic both subjectively and objectively in terms of image noise power spectrum. Conclusion: A methodology has been developed to generate realistic numerical anthropomorphic phantoms using pre-existing radiation therapy data. The GAD models guarantee that generated organs span the statistical distribution of observed radiation therapy patients, according to the training dataset. The methodology enables radiation therapy treatment assessment with multi-modality imaging and a known ground truth, and without patient-dependent bias.

60 APPLIED LIFE SCIENCES↗

A First Look at the Upcoming SISO Space Reference FOM

Simulation is increasingly used in the space domain for several purposes. One example is analysis and engineering, from the mission level down to individual systems and subsystems. Another example is training of space crew and flight controllers. Several distributed simulations have been developed for example for docking vehicles with the ISS and for mission training, in many cases with participants from several nations. Space based scenarios are also used in the "Simulation Exploration Experience", SISO's university outreach program. We have thus realized that there is a need for a distributed simulation interoperability standard for data exchange within the space domain. Based on these experiences, SISO is developing a Space Reference FOM. Members of the product development group come from several countries and contribute experiences from projects within NASA, ESA and other organizations. Participants represent government, academia and industry. The first version will focus on handling of time and space. The Space Reference FOM will provide the following: (i) a flexible positioning system using reference frames for arbitrary bodies in space, (ii) a naming conventions for well known reference frames, (iii) definitions of common time scales, (iv) federation agreements for common types of time management with focus on time stepped simulation, and (v) support for physical entities, such as space vehicles and astronauts. The Space Reference FOM is expected to make collaboration politically, contractually and technically easier. It is also expected to make collaboration easier to manage and extend.

Crues, Edwin↗