Search NASA⌕ Search

SEARCH · Search NASA

Results for “convolution”

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 775 records · Page 43

Deep Learning-Based Dynamic Modeling of Three-Phase Voltage Source Inverters

Inverter-based resource (IBR) models are necessary to analyze modern power system stability and create effective control strategies. Modeling IBRs in converter-rich power systems is crucial, yet challenging due to the lack of commercial information on converter topologies and control parameters. This paper proposes novel convolutional neural network (CNN)–based data-driven techniques for modeling IBRs, addressing adaptability and proprietary concerns without requiring internal system physics knowledge. The proposed method is tested using real grid-tied commercial IBR transient data and demonstrates effectiveness and accuracy. Furthermore, the developed modeling approach is integrated and implemented in the open-source power distribution simulation and analysis tool, GridLAB-D, to illustrate the potentiality of dynamic analysis of large-scale power systems with high IBRs.

deep learning, artificial intelligence↗

JACC.shared: Leveraging HPC Metaprogramming and Performance Portability for Computations That Use Shared Memory GPUs

In this work, we present JACC.shared, a new feature of Julia for ACCelerators (JACC), which is the performanceportable and metaprogramming model of the just-in-time and LLVM-based Julia language. This new feature allows JACC applications to leverage the high-performance computing (HPC) capabilities of high-bandwidth, on-chip GPU memory. Historically, exploiting high-bandwidth, shared-memory GPUs has not been a priority for high-level programming solutions. JACC.shared covers that gap for the first time, thereby providing a highlevel, portable, and easy-to-use solution for programmers to exploit this memory and supporting all current major accelerator architectures. Well-known HPC and AI workloads, such as multi/hyperspectral imaging and AI convolutions, have been used to evaluate JACC.shared on two exascale GPU architectures hosted by some of the most powerful US Department of Energy supercomputers: Perlmutter (NVIDIA A100) and Frontier (AMD MI250X). The performance evaluation reports speedup of up to 3.5× by adding only one line of code to the base codes, thus providing important accelerators in a simple, portable, and transparent way and elevating the programming productivity and performance-portability capabilities for Julia/JACC HPC, AI, and scientific applications.

Valero Lara, Pedro [ORNL] (ORCID:0000000214794310)↗

Autonomous Human Activity Classification From Wearable Multi-Modal Sensors

There has been significant amount of research work on human activity classification relying either on Inertial Measurement Unit (IMU) data or data from static cameras providing a third-person view. There has been relatively less work using wearable cameras, providing first-person or egocentric view, and even fewer approaches combining egocentric video with IMU data. Utilizing only IMU data limits the variety and complexity of the activities that can be detected. For instance, the sitting activity can be detected by IMU data, but it cannot be determined whether the subject has sat on a chair or a sofa, or where the subject is. To perform fine-grained activity classification, and to distinguish between activities that cannot be differentiated by only IMU data, we present an autonomous and robust method using data from both wearable cameras and IMUs. In contrast to convolutional neural network-based approaches, we propose to employ capsule networks to obtain features from egocentric video data. Moreover, Convolutional Long Short Term Memory framework is employed both on egocentric videos and IMU data to capture the temporal aspect of actions. We also propose a genetic algorithm-based approach to autonomously and systematically set various network parameters, rather than using manual settings. Experiments have been conducted to perform 9- and 26-label activity classification, and the proposed method, using autonomously set network parameters, has provided very promising results, achieving overall accuracies of 86.6% and 77.2%, respectively. The proposed approach, combining both modalities, also provides increased accuracy compared to using only egovision data and only IMU data.

47 OTHER INSTRUMENTATION↗

Automated Labeling of Electron Microscopy Images Using Deep Learning

© 2018 IEEE. Searching for scientific data requires metadata providing a relevant context. Today, generating metadata is a time and labor intensive manual process that is often neglected, and important datasets are not accessible through search. We investigate the use of machine learning to generalize metadata from a subset of labeled data, thus increasing the availability of meaningful metadata for search. Specifically, we consider electron microscopy images collected at the National Center for Electron Microscopy (NCEM) at the Lawrence Berkeley National Laboratory (LBNL) and use deep learning to discern characteristics from a small subset of labeled images and transfer labels to the entire image corpus. Relatively small training set sizes and a minimum resolution of 512\times 512 pixels required by the application domain pose unique challenges. We overcome these challenges by using a simple yet powerful convolutional network architecture that limits the number of free parameters to lower the required amount of computational power and reduce the risk of overfitting. We achieve a classification accuracy of approximately 80% in discerning between images recorded in two operating modes of the electron microscope-transmission electron microscopy (TEM) and scanning transmission electron microscopy (STEM). We use transfer learning-i.e., re-using the pre-trained convolution layers from the TEM vs. STEM classification problem-to generalize labels and achieve an accuracy of approximately 70 % despite current experiments being limited to small training data sets. We present these predictions as suggestions to domain scientists through a web-based tool to accelerate the labeling process with the goal of further validating our approach and improving the accuracy of automatically created labels.

Weber, Gunther H↗

Fine-Grained Exploitation of Mixed Precision for Faster CNN Training

As deep convolutional neural networks (CNNs) have become increasingly popular and successful at an ever-widening number of machine learning tasks specialized hardware has become increasingly available for training and deploying them. NVIDIA's recent Volta architecture includes tensor cores which perform a fused operation reduced and mixed precision (16-bit multiply, 32-bit accumulate). Recent research indicates that, typically, very little is lost (in terms of training accuracy) when half precision is used in place of single precision, and performance gains can be made by doing arithmetic in reduced precision. In this work we demonstrate that making layer-by-layer choices as to the arithmetic/data precision can lead to further performance improvement. In our study of 25,200 CNNs we demonstrate an average speedup (over purely half precision) of 1.27x and speedups as high as 3.64x by appropriately combining single and half precision arithmetic and data types on a layer-by-layer basis.

Johnston, Travis↗

Reducing Communication in Graph Neural Network Training

Graph Neural Networks (GNNs) are powerful and flexible neural networks that use the naturally sparse connectivity information of the data. GNNs represent this connectivity as sparse matrices, which have lower arithmetic intensity and thus higher communication costs compared to dense matrices, making GNNs harder to scale to high concurrencies than convolutional or fully-connected neural networks. Here, we introduce a family of parallel algorithms for training GNNs and show that they can asymptotically reduce communication compared to previous parallel GNN training methods. We implement these algorithms, which are based on 1D, 1. 5D, 2D, and 3D sparse-dense matrix multiplication, using torch.distributed on GPU-equipped clusters. Our algorithms optimize communication across the full GNN training pipeline. We train GNNs on over a hundred GPUs on multiple datasets, including a protein network with over a billion edges.

97 MATHEMATICS AND COMPUTING↗

Multiscale Data-Driven Seismic Full-Waveform Inversion With Field Data Study

Seismic full-waveform inversion (FWI), which uses iterative methods to estimate high-resolution subsurface models from seismograms, is a powerful imaging technique in exploration geophysics. In recent years, the computational cost of FWI has grown exponentially due to the increasing size and resolution of seismic data. Moreover, it is a nonconvex problem and can encounter local minima due to the limited accuracy of the initial velocity models or the absence of low frequencies in the measurements. To overcome these computational issues, we develop a multiscale data-driven FWI method based on fully convolutional networks (FCNs). In preparing the training data, we first develop a real-time style transform method to create a large set of synthetic subsurface velocity models from natural images. We then develop two convolutional neural networks with encoder-decoder structures to reconstruct the low- and high-frequency components of the subsurface velocity models, separately. To validate the performance of our data-driven inversion method and the effectiveness of the synthesized training set, we compare it with conventional physics-based waveform inversion approaches using both synthetic and field data. Finally, these numerical results demonstrate that, once our model is fully trained, it can significantly reduce the computation time and yield more accurate subsurface velocity models in comparison with conventional FWI.

58 GEOSCIENCES↗

Iterative Reconstruction of Coded Source Neutron Radiographs

Use of a coded source facilitates high-resolution neutron imaging through magnifications but requires that the radiographic data be deconvolved. A comparison of direct deconvolution with two different iterative algorithms has been performed. One iterative algorithm is based on a maximum likelihood estimation (MLE)-like framework and the second is based on a geometric model of the neutron beam within a least squares formulation of the inverse imaging problem. Simulated data for both uniform and Gaussian shaped source distributions was used for testing to understand the impact of non-uniformities present in neutron beam distributions on the reconstructed images. Results indicate that the model based reconstruction method will match resolution and improve on contrast over convolution methods in the presence of non-uniform sources. Additionally, the model based iterative algorithm provides direct calculation of quantitative transmission values while the convolution based methods must be normalized base on known values.

Santos-Villalobos, Hector J↗

Comparing Mapper Graphs of Artificial Neuron Activations

The mapper graph is a popular tool from topological data analysis that provides a graphical summary of point cloud data. It has been used to study data from cancer research, sports analytics, neurosciences, and machine learning. In particular, mapper graphs have been used recently to visualize the topology of high-dimensional artificial neural activations from convolutional neural networks and large language models. However, a key question that arises from using mapper graphs across applications is how to compare mapper graphs to study their structural differences. In this paper, we introduce a distance between mapper graphs using tools from optimal transport. We demonstrate the utility of such a distance by studying the topological changes of neural activations across convolutional layers in deep learning, as well as by capturing the loss of structural information for multiscale mapper.

mapper graphs, computational topology, machine lea↗

Accelerating Binarized Neural Networks via Bit-Tensor-Cores in Turing GPUs

Despite foreseeing tremendous speedups over conventional deep neural networks, the performance advantage of binarized neural networks (BNNs) has merely been showcased on general-purpose processors such as CPUs and GPUs. In fact, due to being unable to leverage bit-level-parallelism with a word-based architecture, GPUs have been criticized for extremely low utilization (1%) when executing BNNs. Consequently, the latest tensorcores in NVIDIA Turing GPUs start to experimentally support bit computation. In this work, we look into this brand new bit computation capability and characterize its unique features. We show that the stride of memory access can significantly affect performance delivery and a data-format co-design is highly desired to support the tensorcores for achieving superior performance than existing software solutions without tensorcores. We realize the tensorcore-accelerated BNN design, particularly the major functions for fully-connect and convolution layers — bit matrix multiplication and bit convolution. Evaluations on two NVIDIA Turing GPUs show that, with ResNet-18, our BTC-BNN design can process ImageNet at a rate of 5.6K images per second, 77% faster than state-of-the-art. Our BNN approach is released on https://github.com/pnnl/TCBNN.

Li, Ang↗

Frequency Disturbance Event Detection Based on Synchrophasors and Deep Learning

Power system frequency disturbances are caused by various generation and transmission events including generator trips, load disconnections, line trips, etc. Accurate detections of the events are crucial to bulk power system situation awareness and event investigation. This paper utilizes the recent advances of deep learning to build a convolutional neural network model to detect events in an accurate yet straightforward manner. Herein, the rate of change of frequency and the relative angle shift are converted to images as the inputs of the proposed model. Finally, this paper uses two convolutional neural networks and classifier fusion to achieve the detection result. Compared with the conventional event detection algorithm and the frequency only deep learning model, the proposed model improves the detection accuracy by over 48%. As a promising tool for bulk power system situation awareness, the proposed model requires a short decision time, which is suitable for practical scenarios.

42 ENGINEERING↗

Data-Efficient Dimensionality Reduction and Surrogate Modeling of High-Dimensional Stress Fields

Tensor datatypes representing field variables like stress, displacement, velocity, etc., have increasingly become a common occurrence in data-driven modeling and analysis of simulations. Numerous methods [such as convolutional neural networks (CNNs)] exist to address the meta-modeling of field data from simulations. As the complexity of the simulation increases, so does the cost of acquisition, leading to limited data scenarios. Modeling of tensor datatypes under limited data scenarios remains a hindrance for engineering applications. Here, in this article, we introduce a direct image-to-image modeling framework of convolutional autoencoders enhanced by information bottleneck loss function to tackle the tensor data types with limited data. The information bottleneck method penalizes the nuisance information in the latent space while maximizing relevant information making it robust for limited data scenarios. The entire neural network framework is further combined with robust hyperparameter optimization. We perform numerical studies to compare the predictive performance of the proposed method with a dimensionality reduction-based surrogate modeling framework on a representative linear elastic ellipsoidal void problem with uniaxial loading. The data structure focuses on the low-data regime (fewer than 100 data points) and includes the parameterized geometry of the ellipsoidal void as the input and the predicted stress field as the output. The results of the numerical studies show that the information bottleneck approach yields improved overall accuracy and more precise prediction of the extremes of the stress field. Additionally, an in-depth analysis is carried out to elucidate the information compression behavior of the proposed framework.

artificial intelligence↗

In silico simulations of tunneling barrier measurements for molecular orbital-mediated junctions: A molecular orbital theory approach to scanning tunneling microscopy

A new simulator for scanning tunneling microscopy (STM) is presented based on the linear combination of atomic orbitals molecular orbital (LCAO-MO) approximation for the effective tunneling Hamiltonian, which leads to the convolution integral when applied to the tip interaction with the sample. This approach intrinsically includes the structure of the STM tip. Through this mechanical emulation and the tip-inclusive convolution model, dI/dz images for molecular orbitals (which are closely associated with apparent barrier height, ϕ{sub ap}) are reported for the first time. For molecular adsorbates whose experimental topographic images correspond well to isolated-molecule quantum chemistry calculations, the simulator makes accurate predictions, as illustrated by various cases. Distortions in these images due to the tip are shown to be in accord with those observed experimentally and predicted by other ab initio considerations of tip structure. Simulations of the tunneling current dI/dz images are in strong agreement with experiment. The theoretical framework provides a solid foundation which may be applied to LCAO cluster models of adsorbate–substrate systems, and is extendable to emulate several aspects of functional STM operation.

42 ENGINEERING↗

Image transformers for classifying acute lymphoblastic leukemia

Cancer is the leading cause of death by disease in American children. Each year, nearly 16,000 children in the United States and over 300,000 children globally are diagnosed with cancer. Leukemia is a form of blood cancer that originates in the bone marrow and accounts for one-third of pediatric cancers. This disease occurs when the bone marrow contains 20% or more immature white blood cell blasts. Acute lymphoblastic leukemia is the most prevalent leukemia type found in children, with half of all annual cases in the U.S. diagnosed for subjects under 20 years of age. To diagnose acute lymphoblastic leukemia, pathologists often conduct a morphological bone marrow assessment. This assessment determines whether the immature white blood cell blasts in bone marrow display the correct morphological characteristics, such as size and appearance of nuclei. Pathologists also use immunophenotyping via multi-channel flow cytometry to test whether certain antigens are present on the surface of blast cells; the antigens are used to identify the cell lineage of acute lymphoblastic leukemia. These manual processes require well-trained personnel and medical professionals, thus being costly in time and expenses. Computerized decision support via machine learning can accelerate the diagnosis process and reduce the cost. Training a reliable classification model to distinguish between mature and immature white blood cells is essential to the decision support system. Here, we adopted the Vision Transformer model to classify white blood cells. The Vision Transformer achieved superb classification performance compared to state-of-the-art convolutional neural networks while requiring less computational resources for training. Additionally, the latent self-attention architecture provided attention maps for a given image, providing clues as to which portion(s) of the image were significant in decision-making. We applied the Vision Transformer model and a convolutional neural network model to an acute lymphoblastic leukemia classification dataset of 12,528 samples and achieved accuracies of 88.4% and 86.2%.

Cho, Priscilla↗

Differences among Monte Carlo codes in the calculations of voxel S values for radionuclide targeted therapy and analysis of their impact on absorbed dose evaluations

Several updated Monte Carlo (MC) codes are available to perform calculations of voxel S values for radionuclide targeted therapy. The aim of this work is to analyze the differences in the calculations obtained by different MC codes and their impact on absorbed dose evaluations performed by voxel dosimetry. Voxel S values for monoenergetic sources (electrons and photons) and different radionuclides ({sup 90}Y, {sup 131}I, and {sup 188}Re) were calculated. Simulations were performed in soft tissue. Three general-purpose MC codes were employed for simulating radiation transport: MCNP4C, EGSnrc, and GEANT4. The data published by the MIRD Committee in Pamphlet No. 17, obtained with the EGS4 MC code, were also included in the comparisons. The impact of the differences (in terms of voxel S values) among the MC codes was also studied by convolution calculations of the absorbed dose in a volume of interest. For uniform activity distribution of a given radionuclide, dose calculations were performed on spherical and elliptical volumes, varying the mass from 1 to 500 g. For simulations with monochromatic sources, differences for self-irradiation voxel S values were mostly confined within 10% for both photons and electrons, but with electron energy less than 500 keV, the voxel S values referred to the first neighbor voxels showed large differences (up to 130%, with respect to EGSnrc) among the updated MC codes. For radionuclide simulations, noticeable differences arose in voxel S values, especially in the bremsstrahlung tails, or when a high contribution from electrons with energy of less than 500 keV is involved. In particular, for {sup 90}Y the updated codes showed a remarkable divergence in the bremsstrahlung region (up to about 90% in terms of voxel S values) with respect to the EGS4 code. Further, variations were observed up to about 30%, for small source-target voxel distances, when low-energy electrons cover an important part of the emission spectrum of the radionuclide (in our case, for {sup 131}I). For {sup 90}Y and {sup 188}Re, the differences among the various codes have a negligible impact (within few percents) on convolution calculations of the absorbed dose; thus either one of the MC programs is suitable to produce voxel S values for radionuclide targeted therapy dosimetry. However, if a low-energy beta-emitting radionuclide is considered, these differences can affect also dose depositions at small source-target voxel distances, leading to more conspicuous variations (about 9% for {sup 131}I) when calculating the absorbed dose in the volume of interest.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Dosimetric characterization of the iBEAM evo carbon fiber couch for radiotherapy

Purpose: This study characterizes the dosimetric properties of the iBEAM evo carbon fiber couch manufactured by Medical Intelligence and examines the accuracy of the CMS XiO and Nucletron Oncentra Masterplan (OMP) treatment planning systems for calculating beam attenuation due to the presence of the couch. Methods: To assess the homogeneity of the couch, it was CT scanned at isocentric height and a number of signal intensity profiles were generated and analyzed. To simplify experimental procedures, surface dose and central axis depth dose measurements were performed in a solid water slab phantom using Gafchromic film for 6 and 10 MV photon beams at gantry angles of 0 deg. (normal incidence), 30 deg., and 60 deg. with an inverted iBEAM couch placed on top of the phantom. Attenuation measurements were performed in a cylindrical solid water phantom with an ionization chamber positioned at the isocenter. Measurements were taken for gantry angles from 0 deg. to 90 deg. in 10 deg. increments for both 6 and 10 MV photon beams. This setup was replicated in the XiO and OMP treatment planning systems. Dose was calculated using the pencil beam, collapsed cone, convolution, and superposition algorithms. Results: The CT scan of the couch showed that it was uniformly constructed. Surface dose increased by (510{+-}30)% for a 6 MV beam and (600{+-}20)% for a 10 MV beam passing through the couch at normal incidence. Obliquely incident beams resulted in a higher surface dose compared to normally incident beams for both open fields and fields with the couch present. Depth dose curves showed that the presence of the couch resulted in an increase in dose in the build up region. For 6 and 10 MV beams incident at 60 deg., nearly all skin sparing was lost. Attenuation measurements derived using the ionization chamber varied from 2.7% (0 deg.) to a maximum of 4.6% (50 deg.) for a 6 MV beam and from 1.9% (0 deg.) to a maximum of 4.0% (50 deg.) for a 10 MV beam. The pencil beam and convolution algorithms failed to accurately calculate couch attenuation. The collapsed cone and superposition algorithms calculated attenuation within an absolute error of {+-}1.2% for 6 MV and {+-}0.8% for 10 MV for gantry angles from 0 deg. to 40 deg. Some differences in attenuation were observed dependent on how the couch was contoured. Conclusions: These results demonstrate that the presence of the iBEAM evo carbon fiber couch increases the surface dose and dose in the build up region. The inclusion of the couch in the planning scan is limited by the field of view employed and the couch height at the time of CT scanning.

60 APPLIED LIFE SCIENCES↗

Portal dosimetry for VMAT using integrated images obtained during treatment

Purpose: Portal dosimetry provides an accurate and convenient means of verifying dose delivered to the patient. A simple method for carrying out portal dosimetry for volumetric modulated arc therapy (VMAT) is described, together with phantom measurements demonstrating the validity of the approach. Methods: Portal images were predicted by projecting dose in the isocentric plane through to the portal image plane, with exponential attenuation and convolution with a double-Gaussian scatter function. Appropriate parameters for the projection were selected by fitting the calculation model to portal images measured on an iViewGT portal imager (Elekta AB, Stockholm, Sweden) for a variety of phantom thicknesses and field sizes. This model was then used to predict the portal image resulting from each control point of a VMAT arc. Finally, all these control point images were summed to predict the overall integrated portal image for the whole arc. The calculated and measured integrated portal images were compared for three lung and three esophagus plans delivered to a thorax phantom, and three prostate plans delivered to a homogeneous phantom, using a gamma index for 3% and 3 mm. A 0.6 cm{sup 3} ionization chamber was used to verify the planned isocentric dose. The sensitivity of this method to errors in monitor units, field shaping, gantry angle, and phantom position was also evaluated by means of computer simulations. Results: The calculation model for portal dose prediction was able to accurately compute the portal images due to simple square fields delivered to solid water phantoms. The integrated images of VMAT treatments delivered to phantoms were also correctly predicted by the method. The proportion of the images with a gamma index of less than unity was 93.7% ± 3.0% (1SD) and the difference between isocenter dose calculated by the planning system and measured by the ionization chamber was 0.8% ± 1.0%. The method was highly sensitive to errors in monitor units and field shape, but less sensitive to errors in gantry angle or phantom position. Conclusions: This method of predicting integrated portal images provides a convenient means of verifying dose delivered using VMAT, with minimal image acquisition and data processing requirements.

61 RADIATION PROTECTION AND DOSIMETRY↗

Measurement of cardiac output from dynamic pulmonary circulation time CT

Purpose: To introduce a method of estimating cardiac output from the dynamic pulmonary circulation time CT that is primarily used to determine the optimal time window of CT pulmonary angiography (CTPA). Methods: Dynamic pulmonary circulation time CT series, acquired for eight patients, were retrospectively analyzed. The dynamic CT series was acquired, prior to the main CTPA, in cine mode (1 frame/s) for a single slice at the level of the main pulmonary artery covering the cross sections of ascending aorta (AA) and descending aorta (DA) during the infusion of iodinated contrast. The time series of contrast changes obtained for DA, which is the downstream of AA, was assumed to be related to the time series for AA by the convolution with a delay function. The delay time constant in the delay function, representing the average time interval between the cross sections of AA and DA, was determined by least square error fitting between the convoluted AA time series and the DA time series. The cardiac output was then calculated by dividing the volume of the aortic arch between the cross sections of AA and DA (estimated from the single slice CT image) by the average time interval, and multiplying the result by a correction factor. Results: The mean cardiac output value for the six patients was 5.11 (l/min) (with a standard deviation of 1.57 l/min), which is in good agreement with the literature value; the data for the other two patients were too noisy for processing. Conclusions: The dynamic single-slice pulmonary circulation time CT series also can be used to estimate cardiac output.

62 RADIOLOGY AND NUCLEAR MEDICINE↗