Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “NEURAL NETWORKS”

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 577 records · Page 32
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

Scalable balanced training of conditional generative adversarial neural networks on image data

Here, we propose a distributed approach to train deep convolutional generative adversarial neural network (DC-CGANs) models. Our method reduces the imbalance between generator and discriminator by partitioning the training data according to data labels, and enhances scalability by performing a parallel training where multiple generators are concurrently trained, each one of them focusing on a single data label. Performance is assessed in terms of inception score, Fréchet inception distance, and image quality on MNIST, CIFAR10, CIFAR100, and ImageNet1k datasets, showing a significant improvement in comparison to state-of-the-art techniques to training DC-CGANs. Weak scaling is attained on all the four datasets using up to 1000 processes and 2000 NVIDIA V100 GPUs on the OLCF supercomputer Summit.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Surrogate modeling of Monte Carlo radiation transport with convolutional neural networks for shielding optimization

Here, we present a machine learning (ML)-based surrogate model using convolutional neural networks (CNN) designed to emulate the attenuation of neutron fields as they pass through various shielding materials. This model can compute the outgoing neutron flux almost instantaneously and achieves reasonable accuracy compared to traditional Monte Carlo (MC)-based codes, which are computationally intensive. This emulator alleviates the complexity of neutron radiation transport through shielding materials by reducing the dimensionality and enables shielding optimization for a known radiation environment. This optimization process, which would have taken an unrealistic timeline due to several complex radiation transport simulations, can now be achieved in minutes, thus increasing computational capabilities in radiation shielding assessment. We demonstrate the applications of this emulator in computing effective dose rates and optimizing shielding solutions for a heavy-ion accelerator facility, such as the Facility for Rare Isotope Beams, where secondary neutrons produced via beam interactions dominate the radiation environment.

accelerator shielding↗
DOE OSTI2023

Abisko: Deep codesign of an architecture for spiking neural networks using novel neuromorphic materials

The Abisko project aims to develop an energy-efficient spiking neural network (SNN) computing architecture and software system capable of autonomous learning and operation. The SNN architecture explores novel neuromorphic devices that are based on resistive-switching materials, such as memristors and electrochemical RAM. Equally important, Abisko uses a deep codesign approach to pursue this goal by engaging experts from across the entire range of disciplines: materials, devices and circuits, architectures and integration, software, and algorithms. Here, the key objectives of our Abisko project are threefold. First, we are designing an energy-optimized high-performance neuromorphic accelerator based on SNNs. This architecture is being designed as a chiplet that can be deployed in contemporary computer architectures and we are investigating novel neuromorphic materials to improve its design. Second, we are concurrently developing a productive software stack for the neuromorphic accelerator that will also be portable to other architectures, such as field-programmable gate arrays and GPUs. Third, we are creating a new deep codesign methodology and framework for developing clear interfaces, requirements, and metrics between each level of abstraction to enable the system design to be explored and implemented interchangeably with execution, measurement, a model, or simulation. As a motivating application for this codesign effort, we target the use of SNNs for an analog event detector for a high-energy physics sensor.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

Scalable training of graph convolutional neural networks for fast and accurate predictions of HOMO-LUMO gap in molecules

Abstract Graph Convolutional Neural Network (GCNN) is a popular class of deep learning (DL) models in material science to predict material properties from the graph representation of molecular structures. Training an accurate and comprehensive GCNN surrogate for molecular design requires large-scale graph datasets and is usually a time-consuming process. Recent advances in GPUs and distributed computing open a path to reduce the computational cost for GCNN training effectively. However, efficient utilization of high performance computing (HPC) resources for training requires simultaneously optimizing large-scale data management and scalable stochastic batched optimization techniques. In this work, we focus on building GCNN models on HPC systems to predict material properties of millions of molecules. We use HydraGNN, our in-house library for large-scale GCNN training, leveraging distributed data parallelism in PyTorch. We use ADIOS, a high-performance data management framework for efficient storage and reading of large molecular graph data. We perform parallel training on two open-source large-scale graph datasets to build a GCNN predictor for an important quantum property known as the HOMO-LUMO gap. We measure the scalability, accuracy, and convergence of our approach on two DOE supercomputers: the Summit supercomputer at the Oak Ridge Leadership Computing Facility (OLCF) and the Perlmutter system at the National Energy Research Scientific Computing Center (NERSC). We present our experimental results with HydraGNN showing (i) reduction of data loading time up to 4.2 times compared with a conventional method and (ii) linear scaling performance for training up to 1024 GPUs on both Summit and Perlmutter.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2020

Reionization history constraints from neural network based predictions of high-redshift quasar continua

ABSTRACT Observations of the early Universe suggest that reionization was complete by z ∼ 6, however, the exact history of this process is still unknown. One method for measuring the evolution of the neutral fraction throughout this epoch is via observing the Lyα damping wings of high-redshift quasars. In order to constrain the neutral fraction from quasar observations, one needs an accurate model of the quasar spectrum around Lyα, after the spectrum has been processed by its host galaxy but before it is altered by absorption and damping in the intervening intergalactic medium (IGM). In this paper, we present a novel machine learning approach, using artificial neural networks, to reconstruct quasar continua around Lyα. Our Quasar Spectra from Artificial Neural Network based predictive Regression Algorithm(QSANNdRA) improves the error in this reconstruction compared to the state-of-the-art principal component analysis (PCA) based model in the literature by 14.2 per cent on average, and provides an improvement of 6.1 per cent on average when compared to an extension thereof. In comparison with the extended PCA model, QSANNdRA further achieves an improvement of 22.1 per cent and 16.8 per cent when evaluated on low-redshift quasars most similar to the two high-redshift quasars under consideration, ULAS J1120+0641 at z = 7.0851 and ULAS J1342+0928 at z = 7.5413, respectively. Using our more accurate reconstructions of these two z > 7 quasars, we estimate the neutral fraction of the IGM using a homogeneous reionization model and find $\bar{x}_\mathrm{H\, \small{I}} = 0.25^{+0.05}_{-0.05}$ at z = 7.0851 and $\bar{x}_\mathrm{H\, \small{I}} = 0.60^{+0.11}_{-0.11}$ at z = 7.5413. Our results are consistent with the literature and favour a rapid end to reionization.

Ďurovčíková, Dominika↗
DOE OSTI2025

CONCURRENT, CONDENSED STEIN VARIATIONAL GRADIENT DESCENT FOR UNCERTAINTY QUANTIFICATION OF NEURAL NETWORKS

In this work, we propose a Stein variational gradient descent (SVGD) method to concurrently sparsify, train, and provide uncertainty quantification (UQ) of a complexly parameterized model, such as a neural network (NN). It employs a graph reconciliation and condensation process to reduce complexity and increase similarity in the Stein ensemble of parameterizations. Therefore, the proposed concurrent, condensed SVGD (ccSVGD) method can provide UQ on parameters, not just outputs. Furthermore, the parameter reduction speeds up the convergence of the Stein gradient descent as it reduces the combinatorial complexity by aligning and differentiating the sensitivity to parameters. These properties are demonstrated with an illustrative example and an application to a mechanical response representation problem in solid mechanics.

42 ENGINEERING↗
DOE OSTI2021

Extraction of the muon signals recorded with the surface detector of the Pierre Auger Observatory using recurrent neural networks

The Pierre Auger Observatory, at present the largest cosmic-ray observatory ever built, is instrumented with a ground array of 1600 water-Cherenkov detectors, known as the Surface Detector (SD). The SD samples the secondary particle content (mostly photons, electrons, positrons and muons) of extensive air showers initiated by cosmic rays with energies ranging from 1017eV up to more than 1020eV. Measuring the independent contribution of the muon component to the total registered signal is crucial to enhance the capability of the Observatory to estimate the mass of the cosmic rays on an event-by-event basis. However, with the current design of the SD, it is difficult to straightforwardly separate the contributions of muons to the SD time traces from those of photons, electrons and positrons. In this paper, we present a method aimed at extracting the muon component of the time traces registered with each individual detector of the SD using Recurrent Neural Networks. We derive the performances of the method by training the neural network on simulations, in which the muon and the electromagnetic components of the traces are known. We conclude this work showing the performance of this method on experimental data of the Pierre Auger Observatory. We find that our predictions agree with the parameterizations obtained by the AGASA collaboration to describe the lateral distributions of the electromagnetic and muonic components of extensive air showers.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2024

Accelerate microstructure evolution simulation using graph neural networks with adaptive spatiotemporal resolution

Abstract Surrogate models driven by sizeable datasets and scientific machine-learning methods have emerged as an attractive microstructure simulation tool with the potential to deliver predictive microstructure evolution dynamics with huge savings in computational costs. Taking 2D and 3D grain growth simulations as an example, we present a completely overhauled computational framework based on graph neural networks with not only excellent agreement to both the ground truth phase-field methods and theoretical predictions, but enhanced accuracy and efficiency compared to previous works based on convolutional neural networks. These improvements can be attributed to the graph representation, both improved predictive power and a more flexible data structure amenable to adaptive mesh refinement. As the simulated microstructures coarsen, our method can adaptively adopt remeshed grids and larger timesteps to achieve further speedup. The data-to-model pipeline with training procedures together with the source codes are provided.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Automated Stellar Spectra Classification with Ensemble Convolutional Neural Network

Large sky survey telescopes have produced a tremendous amount of astronomical data, including spectra. Machine learning methods must be employed to automatically process the spectral data obtained by these telescopes. Classification of stellar spectra by applying deep learning is an important research direction for the automatic classification of high-dimensional celestial spectra. In this paper, a robust ensemble convolutional neural network (ECNN) was designed and applied to improve the classification accuracy of massive stellar spectra from the Sloan digital sky survey. We designed six classifiers which consist six different convolutional neural networks (CNN), respectively, to recognize the spectra in DR16. Then, according the cross-entropy testing error of the spectra at different signal-to-noise ratios, we integrate the results of different classifiers in an ensemble learning way to improve the effect of classification. The experimental result proved that our one-dimensional ECNN strategy could achieve 95.0% accuracy in the classification task of the stellar spectra, a level of accuracy that exceeds that of the classical principal component analysis and support vector machine model.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2024

Survey of gravitationally lensed objects in HSC imaging (SuGOHI) – X. Strong lens finding in the HSC-SSP using convolutional neural networks

ABSTRACT We apply a novel model based on convolutional neural networks (CNN) to identify gravitationally lensed galaxies in multiband imaging of the Hyper Suprime-Cam Subaru Strategic Program (HSC-SSP) Survey. The trained model is applied to a parent sample of 2350 061 galaxies selected from the $\sim$ 800 deg$^2$ Wide area of the HSC-SSP Public Data Release 2. The galaxies in HSC Wide are selected based on stringent pre-selection criteria, such as multiband magnitudes, stellar mass, star formation rate, extendedness limit, photometric redshift range, etc. The trained CNN assigns a score from 0 to 1, with 1 representing lenses and 0 representing non-lenses. Initially, the CNN selects a total of 20 241 cutouts with a score greater than 0.9, but this number is subsequently reduced to 1522 cutouts after removing definite non-lenses for further visual inspection. We discover 43 grade A (definite) and 269 grade B (probable) strong lens candidates, of which 97 are completely new. In addition, we also discover 880 grade C (possible) lens candidates, 289 of which are known systems in the literature. We identify 143 candidates from the known systems of grade C that had higher confidence in previous searches. Our model can also recover 285 candidate galaxy-scale lenses from the Survey of Gravitationally lensed Objects in HSC Imaging (SuGOHI), where a single foreground galaxy acts as the deflector. Even though group-scale and cluster-scale lens systems are not included in the training, a sample of 32 SuGOHI-c (i.e. group/cluster-scale systems) lens candidates is retrieved. Our discoveries will be useful for ongoing and planned spectroscopic surveys, such as the Subaru Prime Focus Spectrograph project, to measure lens and source redshifts in order to enable detailed lens modelling.

Jaelani, Anton T. (ORCID:0000000162825778)↗
DOE OSTI2025

GSplit: Scaling Graph Neural Network Training on Large Graphs via Split-Parallelism

Graph neural networks (GNNs), an emerging class of machine learning models for graphs, have gained popularity for their superior performance in various graph analytical tasks. Mini-batch training is commonly used to train GNNs on large graphs, and data parallelism is the standard approach to scale mini-batch training across multiple GPUs. Data parallel approaches contain redundant work as subgraphs sampled by different GPUs contain significant overlap. To address this issue, we introduce a hybrid parallel mini-batch training paradigm called Split parallelism. Split parallelism avoids redundant work by splitting the sampling, loading, and training of each mini-batch across multiple GPUs. Split parallelism, however, introduces communication overheads that can be more than the savings from removing redundant work. We further present a lightweight partitioning algorithm that probabilistically minimizes these overheads. We implement spllit parllelism in GSplit and show that it outperforms state-of-the-art mini-batch training systems like DGL, Quiver, and P3.

Lim, Seung-Hwan [ORNL] (ORCID:0000000194616866)↗
DOE OSTI2025

A Comparative Study of Physics‐Informed and Data‐Driven Neural Networks for Compound Flood Simulation at River‐Ocean Interfaces: A Case Study of Hurricane Irene

Simulating compound flooding (CF) at the river-ocean interface within large-scale Earth System Models (ESMs) presents significant challenges due to complex interactions between river discharge, storm surge, and tides. This study assesses the comparative advantages of physics-informed and data-driven machine learning (ML) approaches for enhancing local ESM performance. We systematically compare data-driven neural network models (i.e., CNNs, U-Net, Long Short-Term Memory (LSTM), Gated Recurrent Unit), and physics-informed neural network (PINN) models, including vanilla PINN and a finite-difference-based PINN (FD-PINN). Specifically, FD-PINN is introduced to enhance computational efficiency, accelerating vanilla PINNs by ∼6.5 times while improving accuracy. To enhance data-driven model training, a new data-generation approach is developed to sample historical fluvial and coastal flood events, which ensures a robust data set for extreme event prediction. The models are evaluated using a realistic one-dimensional river domain extracted from an ESM's river mesh and the Hurricane Irene event as an independent test case. Results show that FD-PINN achieves accurate predictions with significantly reduced computational costs relative to vanilla PINNs. Among data-driven models, the best overall performance is achieved by a CNN-LSTM hybrid, which balances accuracy and efficiency. While a fully connected CNN (CNN-FC) provides the best accuracy, it incurs high computational cost. Architectures lacking strong temporal modeling tend to underperform on unseen events. These findings highlight the importance of sequence-aware designs for robust generalization. This study reveals the trade-offs between physics-informed and data-driven models and proposes an adaptive hybrid framework for integrating ML into ESMs to enhance local flood simulations.

Earth Systems Modeling↗
DOE OSTI2025

Synthetic-domain computing and neural networks using lithium niobate integrated nonlinear phononics

Analogue computing uses the physical behaviours of devices to provide energy-efficient arithmetic operations. However, scaling up analogue computing platforms by simply increasing the number of devices leads to challenges such as device-to-device variation. Here, in this study, we report scalable analogue computing and neural networks in the synthetic frequency domain using an integrated nonlinear phononic platform on lithium niobate. This synthetic-domain computing is robust to device variations, as vectors and matrices are concurrently encoded at different frequencies within a single device, achieving a high throughput per area. Leveraging inherent nonlinearities, our device-aware neural network can perform a four-class classification task with an accuracy of 98.2%. The nonlinear phononic computing hardware also maintains consistent performance over a wide operational temperature range (characterized up to 192 °C). Our synthetic-domain computing combines single-device parallelism, inherent nonlinearity and environmental stability, and could be of use in edge computing applications in which power efficiency and environmental resilience are crucial.

Ji, Jun [Virginia Polytechnic Inst. and State Univ↗
DOE OSTI2024

Neural network kinetics for exploring diffusion multiplicity and chemical ordering in compositionally complex materials

Diffusion involving atom transport from one location to another governs many important processes and behaviors such as precipitation and phase nucleation. The inherent chemical complexity in compositionally complex materials poses challenges for modeling atomic diffusion and the resulting formation of chemically ordered structures. Here, we introduce a neural network kinetics (NNK) scheme that predicts and simulates diffusion-induced chemical and structural evolution in complex concentrated chemical environments. The framework is grounded on efficient on-lattice structure and chemistry representation combined with artificial neural networks, enabling precise prediction of all path-dependent migration barriers and individual atom jumps. To demonstrate the method, we study the temperature-dependent local chemical ordering in a refractory NbMoTa alloy and reveal a critical temperature at which the B2 order reaches a maximum. The atomic jump randomness map exhibits the highest diffusion heterogeneity (multiplicity) in the vicinity of this characteristic temperature, which is closely related to chemical ordering and B2 structure formation. The scalable NNK framework provides a promising new avenue to exploring diffusion-related properties in the vast compositional space within which extraordinary properties are hidden.

36 MATERIALS SCIENCE↗
DOE OSTI2022

A neural network‐enhanced reproducing kernel particle method for modeling strain localization

Abstract Modeling the localized intensive deformation in a damaged solid requires highly refined discretization for accurate prediction, which significantly increases the computational cost. Although adaptive model refinement can be employed for enhanced effectiveness, it is cumbersome for the traditional mesh‐based methods to perform while modeling the evolving localizations. In this work, neural network‐enhanced reproducing kernel particle method (NN‐RKPM) is proposed, where the location, orientation, and shape of the solution transition near a localization is automatically captured by the NN approximation via a block‐level neural network (NN) optimization. The weights and biases in the blocked parameterization network control the location and orientation of the localization. The designed basic four‐kernel NN block is capable of capturing a triple junction or a quadruple junction topological pattern, while more complicated localization topological patters are captured by the superposition of multiple four‐kernel NN blocks. The standard RK approximation is then utilized to approximate the smooth part of the solution, which permits a much coarser discretization than the high‐resolution discretization needed to capture sharp solution transitions with the conventional methods. A regularization of the NN approximation is additionally introduced for discretization‐independent material responses. The effectiveness of the proposed NN‐RKPM is verified by a series of numerical verifications.

Baek, Jonghyuk↗
DOE OSTI2025

Rapid Inference of Logic Gate Neural Networks for Anomaly Detection in High Energy Physics

The increasing data rates and complexity of detectors at the Large Hadron Collider (LHC) necessitate fast and efficient machine learning models, particularly for rapid selection of what data to store, known as triggering. Building on recent work in differentiable logic gates, we present a public implementation of a Convolutional Differentiable Logic Gate Neural Network (CLGN). We apply this to detecting anomalies at the Level-1 Trigger at CMS using public data from the CICADA project. We demonstrate that the CLGN achieves physics performance on par with or superior to conventional quantized neural networks. We also synthesize an LGN for a Field-Programmable Gate Array (FPGA) and show highly promising FPGA characteristics, notably zero Digital Signal Processor (DSP) resource usage. This work highlights the potential of logic gate networks for high-speed, on-detector inference in High Energy Physics and beyond.

FOS: Physical sciences↗
DOE OSTI2022

Physics-informed neural networks for identification of material properties using standing waves

A metallic structure in its initial stage of failure involves plastic deformation or environmental degradation that changes the elastic modulus and density. This work presents the detection of change in wave velocity (a function of elastic modulus and density) as a system identification problem. A physics-informed neural network (PINN) is proposed to solve the system identification problem. The PINN takes the spatial coordinates of scanning locations and time as inputs and provides the displacement and wave velocity as outputs. The governing partial differential equation of standing waves in a rod is incorporated into the neural network as physics in the form of a loss function. The wave velocity vector is randomly initiated. During the training of the network, physics is used to determine and update the wave velocity target vector from the network’s displacement predictions. The measured data, comprising sparse displacement response on the rod structure, are used to train the PINN. The wave velocity at the sparse locations on the rod is learned from the predicted displacements during the training. Using the predictions of the trained network, the response of free vibration or material property variation can be reconstructed at unscanned locations on the structure to obtain high-resolution maps for full-field imaging to detect and localize the changes caused by plastic deformation. The PINN’s sparse scanning and simultaneous prediction capability during training can lead to high scanning and data-processing speeds. This capability yields a nondestructive evaluation system that can predict the presence of degraded material locations as the structural vibrations are scanned and processed in real time.

Rathod, Vivek↗