Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Convolutional 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 1,567 records · Page 87
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Efficient Super-Resolution of Near-Surface Climate Modeling Using the Fourier Neural Operator

Downscaling methods are critical in efficiently generating high-resolution atmospheric data. However, state-of-the-art statistical or dynamical downscaling techniques either suffer from the high computational cost of running a physical model or require high-resolution data to develop a downscaling tool. Here, we demonstrate a recently proposed zero-shot super-resolution method, the Fourier neural operator (FNO), to efficiently perform downscaling without the need for high-resolution data. Because the FNO learns dynamics in Fourier space, FNO is a resolution-invariant emulator; it can be trained at a coarse resolution and produces emulation at any high resolution. We applied FNO to downscale a 4-km resolution Weather Research and Forecasting (WRF) Model simulation of near-surface heat-related variables over the Great Lakes region. The FNO is driven by the atmospheric forcings, and topographic features used in the WRF model at the same resolution. We incorporated a physics-constrained loss in FNO by using the Clausius-Clapeyron relation to better constrain the relations among the emulated states. Trained on merely 600 WRF snapshots at 4-km resolution, the FNO shows comparable performance with a widely used convolutional network, U-Net, achieving averaged modified Kling-Gupta Efficiency of 0.88 and 0.94 on the test dataset for temperature and pressure, respectively. We then employed the FNO to produce 1-km emulations to reproduce the fine climate features. Further, by taking the WRF simulation as ground truth, we show consistent performances at the two resolutions, suggesting the reliability of FNO in producing high-resolution dynamics. Our study demonstrates the potential of using FNO for zero-shot super-resolution in generating first-order estimation on atmospheric modeling.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

Adversarial methods to reduce simulation bias in neutrino interaction event filtering at liquid argon time projection chambers

For current and future neutrino oscillation experiments using large liquid argon time projection chambers (LAr-TPCs), a key challenge is identifying neutrino interactions from the pervading cosmic-ray background. Rejection of such background is often possible using traditional cut-based selections, but this typically requires the prior use of computationally expensive reconstruction algorithms. This work demonstrates an alternative approach of using a 3D submanifold sparse convolutional network trained on low-level information from the scintillation light signal of interactions inside LAr-TPCs. This technique is applied to example simulations from ICARUS, the far detector of the short baseline neutrino program at Fermilab. The results of the network, show that cosmic background is reduced by up to 76.3% whilst neutrino interaction selection efficiency remains over 98.9%. We further present a way to mitigate potential biases from imperfect input simulations by applying domain adversarial neural networks (DANNs), for which modified simulated samples are introduced to imitate real data and a small portion of them are used for adversarial training. A series of mock-data studies are performed and demonstrate the effectiveness of using DANNs to mitigate biases, showing neutrino interaction selection efficiency performances significantly better than that achieved without the adversarial training.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2025

Uncovering Structure–Conductivity Relationships in Anion Exchange Membranes (AEMs) Using Interpretable Machine Learning

Anion exchange membranes (AEMs) play a vital role in the performance of water electrolyzers and fuel cells, yet their discovery and optimization remain challenging due to the complexity of structure–property relationships. In this study, we introduce a machine learning framework that leverages conditional graph neural networks (cGNNs) and descriptor-based models and a hybrid graph neural network (HGARE) to predict and interpret ionic conductivity. The descriptor-based pipeline employs principal component analysis (PCA), ablation, and SHAP analysis to identify factors governing anion conductivity, revealing electronic, topological, and compositional descriptors as key contributors. Beyond prediction, dimensionality reduction and clustering are performed by employing t-SNE and KMeans as well as SOM, which reveal distinct membranes clusters, some of which were enriched with high anion conductivity. Among graph-based approaches, the graph convolutional (GCN) achieved strong predictive performance, while the Hybrid Graph Autoencoder-Regressor Ensemble (HGARE) achieved the highest accuracy. Additionally, atom-level saliency maps from GCN provide spatial explanations for conductive behavior, revealing the importance of polarizable and flexible regions. This work contributes to the accelerated and data-driven design of high-performance AEMs.

Naghshnejad, Pegah [Department of Chemical Enginee↗
DOE OSTI2019

An Efficient Bayesian Method for Advancing the Application of Deep Learning in Earth Science

Despite their wide and successful applications, deep learning (DL) models are prone to overfitting for small training datasets, produce a poor predictive performance for uncertain data, and provide point estimations without any indication of the accuracy and credibility. These limitations of the deterministic DL models hinder their effective application in Earth system science where the labelled data are sparse, noisy and incomplete with large uncertainty and where the predictive uncertainty quantification is needed for scientific understanding and policy decision making. Integration of Bayesian inference into DL models adds an estimate of uncertainty and regularization in the predictions. However, traditional Bayesian methods are computationally unaffordable and inflexible for high-dimensional problems, which limits their application to DL systems that typically have millions of model parameters. In this effort, we propose an efficient and general-purpose Bayesian inference method to advance DL model optimization and uncertainty quantification, so as to facilitate the adoption of DL in Earth sciences. In a demonstration, we integrate the proposed Bayesian method with a feedforward neural network (NN) to build a fast-to-evaluate surrogate of the complex Energy Exascale Earth System Land Model for efficient modeling. The formulated Bayesian NN, using a small number of training data, produces an accurate prediction with high credibility, whereas with the same small training size, the deterministic NN cannot yield a reasonable estimation and does not provide confidence information. The proposed Bayesian method is computationally efficient and flexible, capable of integration with diverse network variants such as convolutional NNs and recurrent NNs to advance the application of DL in Earth sciences.

Lu, Dan↗
DOE OSTI2023

CEGANN: CRYSTAL EDGE GRAPH ATTENTION NEURAL NETWORK

SF-22-156 Machine learning (ML) models and applications in materials design and discovery typically involve the use of feature representations or descriptors followed by a learning algorithm that maps them to user desired properties of interest. Most popular mathematical formulation-based descriptors are not unique across atomic environments and suffer from transferability issues across different application domains and/or material classes. The CEGANN code provides a unified interface to facilitate material characterization across materials across multiple scales (from atomic to mesoscale) and diverse classes of materials ranging from metals oxides, non-metals, and even hierarchical materials such as zeolites and semi ordered materials such as mesophases. CEGANN implements a Graph Attention Network (GAT) type convolution architecture. The details of network architecture can be found in the paper https://doi.org/10.48550/arXiv.2207.10168. The software comes with pretrained examples and dataset for the classification of the following representative systems: (1) Structure-level representation such as space group (2) Structural dimensionality (e.g., bulk, 2D, clusters etc.) (3) Grain boundary identification (4) Nucleation and growth of a zeolite polymorph (5) Characterization of binary mesophases and their phase transitions (6) Growth of ice. The code is written in python programming language.

CHAN, HENRYT↗
DOE OSTI2023

A fine pore-preserved deep neural network for porosity analytics of a high burnup U-10Zr metallic fuel

Abstract U-10 wt.% Zr (U-10Zr) metallic fuel is the leading candidate for next-generation sodium-cooled fast reactors. Porosity is one of the most important factors that impacts the performance of U-10Zr metallic fuel. The pores generated by the fission gas accumulation can lead to changes in thermal conductivity, fuel swelling, Fuel-Cladding Chemical Interaction (FCCI) and Fuel-Cladding Mechanical Interaction (FCMI). Therefore, it is crucial to accurately segment and analyze porosity to understand the U-10Zr fuel system to design future fast reactors. To address the above issues, we introduce a workflow to process and analyze multi-source Scanning Electron Microscope (SEM) image data. Moreover, an encoder-decoder-based, deep fully convolutional network is proposed to segment pores accurately by integrating the residual unit and the densely-connected units. Two SEM 250 × field of view image datasets with different formats are utilized to evaluate the new proposed model’s performance. Sufficient comparison results demonstrate that our method quantitatively outperforms two popular deep fully convolutional networks. Furthermore, we conducted experiments on the third SEM 2500 × field of view image dataset, and the transfer learning results show the potential capability to transfer the knowledge from low-magnification images to high-magnification images. Finally, we use a pre-trained network to predict the pores of SEM images in the whole cross-sectional image and obtain quantitative porosity analysis. Our findings will guide the SEM microscopy data collection efficiently, provide a mechanistic understanding of the U-10Zr fuel system and bridge the gap between advanced characterization to fuel system design.

36 MATERIALS SCIENCE↗
DOE OSTI2023

Defect graph neural networks for materials discovery in high-temperature clean-energy applications

We present a graph neural network approach that fully automates the prediction of defect formation enthalpies for any crystallographic site from the ideal crystal structure, without the need to create defected atomic structure models as input. Here we used density functional theory reference data for vacancy defects in oxides, to train a defect graph neural network (dGNN) model that replaces the density functional theory supercell relaxations otherwise required for each symmetrically unique crystal site. Interfaced with thermodynamic calculations of reduction entropies and associated free energies, the dGNN model is applied to the screening of oxides in the Materials Project database, connecting the zero-kelvin defect enthalpies to high-temperature process conditions relevant for solar thermochemical hydrogen production and other energy applications. The dGNN approach is applicable to arbitrary structures with an accuracy limited principally by the amount and diversity of the training data, and it is generalizable to other defect types and advanced graph convolution architectures. In conclusion, it will help to tackle future materials discovery problems in clean energy and beyond.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2018

Sparse coding of pathology slides compared to transfer learning with deep neural networks

Background Histopathology images of tumor biopsies present unique challenges for applying machine learning to the diagnosis and treatment of cancer. The pathology slides are high resolution, often exceeding 1GB, have non-uniform dimensions, and often contain multiple tissue slices of varying sizes surrounded by large empty regions. The locations of abnormal or cancerous cells, which may constitute a small portion of any given tissue sample, are not annotated. Cancer image datasets are also extremely imbalanced, with most slides being associated with relatively common cancers. Since deep representations trained on natural photographs are unlikely to be optimal for classifying pathology slide images, which have different spectral ranges and spatial structure, we here describe an approach for learning features and inferring representations of cancer pathology slides based on sparse coding. Results We show that conventional transfer learning using a state-of-the-art deep learning architecture pre-trained on ImageNet (RESNET) and fine tuned for a binary tumor/no-tumor classification task achieved between 85% and 86% accuracy. However, when all layers up to the last convolutional layer in RESNET are replaced with a single feature map inferred via a sparse coding using a dictionary optimized for sparse reconstruction of unlabeled pathology slides, classification performance improves to over 93%, corresponding to a 54% error reduction. Conclusions We conclude that a feature dictionary optimized for biomedical imagery may in general support better classification performance than does conventional transfer learning using a dictionary pre-trained on natural images.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2019

LP-BNN: Ultra-low-Latency BNN Inference with Layer Parallelism

High inference latency seriously limits the deployment of DNNs in real-time domains such as autonomous driving, robotic control, and many others. To address this emerging challenge, researchers have proposed approximate DNNs with reduced precision, e.g., Binarized Neural Networks (BNNs). While BNNs can be built to have little loss in accuracy, latency reduction still has much room for improvement. In this paper, we propose a single-FPGA-based BNN accelerator that achieves microsecond-level ultra-low-latency inference of ImageNet, LP- BNN. We obtain this performance via several design optimizations. First, we optimize the network structure by removing Batch Normalization (BN) functions which leads to significant latency in BNNs without any loss on accuracy. Second, we propose a parameterized architecture which is based on layer parallelism and supports nearly perfect load balancing for various types of BNNs. Third, we fuse all the convolution layers and the first fully connected layer. We process them in parallel through fine- grained inter-layer pipelining. With our proposed accelerator, the inference of binarized AlexNet, VGGNet, and ResNet are completed within 21.5µs, 355µs, and 67.8µs respectively, with no loss in accuracy as compared with other BNN implementations.

Geng, Tong↗
DOE OSTI2023

ARMing the Edge: Designing Edge Computing–Capable Machine Learning Algorithms to Target ARM Doppler Lidar Processing

Abstract There is a need for long-term observations of cloud and precipitation fall speeds in validating and improving rainfall forecasts from climate models. To this end, the U.S. Department of Energy Atmospheric Radiation Measurement (ARM) user facility Southern Great Plains (SGP) site at Lamont, Oklahoma, hosts five ARM Doppler lidars that can measure cloud and aerosol properties. In particular, the ARM Doppler lidars record Doppler spectra that contain information about the fall speeds of cloud and precipitation particles. However, due to bandwidth and storage constraints, the Doppler spectra are not routinely stored. This calls for the automation of cloud and rain detection in ARM Doppler lidar data so that the spectral data in clouds can be selectively saved and further analyzed. During the ARMing the Edge field experiment, a Waggle node capable of performing machine learning applications in situ was deployed at the ARM SGP site for this purpose. In this paper, we develop and test four algorithms for the Waggle node to automatically classify ARM Doppler lidar data. We demonstrate that supervised learning using a ResNet50-based classifier will classify 97.6% of the clear-air images and 94.7% of cloudy images correctly, outperforming traditional peak detection methods. We also show that a convolutional autoencoder paired with k -means clustering identifies 10 clusters in the ARM Doppler lidar data. Three clusters correspond to mostly clear conditions with scattered high clouds, and seven others correspond to cloudy conditions with varying cloud-base heights.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2019

Fast neural network training on a cluster of GPUs for action recognition with high accuracy

In this work, we propose algorithms and techniques to accelerate training of deep neural networks for action recognition on a cluster of GPUs. The convergence analysis of our algorithm shows it is possible to reduce communication cost and at the same time minimize the number of iterations needed for convergence. We customize the Adam optimizer for our distributed algorithm to improve efficiency. In addition, we employ transfer-learning to further reduce training time while improving validation accuracy. For the UCF101 and HMDB51 datasets, the validation accuracies achieved are 93.1% and 67.9% respectively. With an additional end-to-end trained temporal stream, the validation accuracies achieved for UCF101 and HMDB51 are 93.47% and 81.24% respectively. As far as we know, these are the highest accuracies achieved with the two-stream approach using ResNet that does not involve computationally expensive 3D convolutions or pretraining on much larger datasets.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

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↗
DOE OSTI2018

SuperNeurons: Dynamic GPU Memory Management for Training Deep Neural Networks

Going deeper and wider in neural architectures improves their accuracy, while the limited GPU DRAM places an undesired restriction on the network design domain. Deep Learning (DL) practitioners either need to change to less de- sired network architectures, or nontrivially dissect a network across multiGPUs. These distract DL practitioners from concentrating on their original machine learning tasks. We present SuperNeurons: a dynamic GPU memory scheduling runtime to enable the network training far beyond the GPU DRAM capacity. SuperNeurons features 3 memory optimizations, Liveness Analysis, Unified Tensor Pool, and Cost-Aware Recomputation; together they effectively reduce the network-wide peak memory usage down to the maximal memory usage among layers. We also address the performance issues in these memory-saving techniques. Given the limited GPU DRAM, SuperNeurons not only provisions the necessary memory for the training, but also dynamically allocates the memory for convolution workspaces to achieve the high performance. Evaluations against Caffe, Torch, MXNet and TensorFlow have demonstrated that SuperNeurons trains at least 3.2432 deeper network than current ones with the leading performance. Particularly, SuperNeurons can train ResNet2500 that has 104 basic network layers on a 12GB K40c.

Wang, Linnan↗
DOE OSTI2025

Structural differences between human and mouse neurons and their implementation in generative AIs

Mouse and human brains have different functions that depend on their neuronal networks. We analyzed nanometer-scale three-dimensional structures of brain tissues of the mouse medial prefrontal cortex and compared them with structures of the human anterior cingulate cortex. The obtained results indicated that mouse neuronal somata are smaller and neurites are thinner than those of human neurons. We implemented these characteristics of mouse neurons in convolutional layers of a generative adversarial network (GAN) and a denoising diffusion implicit model (DDIM), which were then subjected to image generation tasks using photo datasets of cat faces, cheese, human faces, birds, and automobiles. The mouse-mimetic GAN outperformed a standard GAN in the image generation task using the cat faces and cheese photo datasets, but underperformed for human faces and birds. The mouse-mimetic DDIM gave similar results, suggesting that the nature of the datasets affected the results. Analyses of the five datasets indicated differences in their image entropy, which should influence the number of parameters required for image generation. The preferences of the mouse-mimetic AIs coincided with the impressions commonly associated with mice. The relationship between the neuronal network and brain function should be investigated by implementing other biological findings in artificial neural networks.

generative AI↗
DOE OSTI2022

High Velocity, Low‐Voltage Collective In‐Plane Switching in (100) BaTiO 3 Thin Films

Abstract Ferroelectrics are being increasingly called upon for electronic devices in extreme environments. Device performance and energy efficiency is highly correlated to clock frequency, operational voltage, and resistive loss. To increase performance it is common to engineer ferroelectric domain structure with highly‐correlated electrical and elastic coupling that elicit fast and efficient collective switching. Designing domain structures with advantageous properties is difficult because the mechanisms involved in collective switching are poorly understood and difficult to investigate. Collective switching is a hierarchical process where the nano‐ and mesoscale responses control the macroscopic properties. Using chemical solution synthesis, epitaxially nearly‐relaxed (100) BaTiO 3 films are synthesized. Thermal strain induces a strongly‐correlated domain structure with alternating domains of polarization along the [010] and [001] in‐plane axes and 90° domain walls along the [011] or [01] directions. Simultaneous capacitance–voltage measurements and band‐excitation piezoresponse force microscopy revealed strong collective switching behavior. Using a deep convolutional autoencoder, hierarchical switching is automatically tracked and the switching pathway is identified. The collective switching velocities are calculated to be ≈500 cm s −1 at 5 V (7 kV cm −1 ), orders‐of‐magnitude faster than expected. These combinations of properties are promising for high‐speed tunable dielectrics and low‐voltage ferroelectric memories and logic.

36 MATERIALS SCIENCE↗
DOE OSTI2023

Attention-Augmented Parametric Kernel Graph Neural Network (APKGNN) for Node Classification

We present a new graph neural network, the Attention-based Parametric-Kernel augmented Graph Neural Network (APKGNN), developed for node classification tasks. Despite extensive work on modeling multi-faceted relationships between connected nodes of a graph, the effect of attention on edge features mapped to relationships has not yet been analyzed through learning representation. This study derives such an attention vector by first calculating node features corresponding to endpoints of an edge and then aggregating these with extracted local intrinsic patches of a given graph to generate augmented local patch vectors. This process uses a parametric kernel based on Gaussian mixture models (GMMs) to embed local neighborhoods of the graph in local patches. The patch vectors then convolve with the above node features to produce an updated node representation. We show that this new learning representation (APKGNN) achieves higher node classification accuracy on tasks - both standard benchmarks (Cora, PubMed, Citeseer) and new experimental short text corpora where nodes correspond to text documents and words. This implementation of the GNN convolution layer outperforms state-of-the-art (SOTA) algorithms, achieving higher training, validation, and test accuracy by a significant margin on three standard benchmark data sets under both SOTA experimental settings and those for new testbeds.

Bose, Avishek↗
DOE OSTI2024

Graph neural network for neutrino physics event reconstruction

Liquid argon time projection chamber (LArTPC) detector technology offers a wealth of high-resolution information on particle interactions, and leveraging that information to its full potential requires sophisticated automated reconstruction techniques. Here, this article describes NUGRAPH 2, a graph neural network for low-level reconstruction of simulated neutrino interactions in a LArTPC detector. Simulated neutrino interactions in the MicroBooNE detector geometry are described as heterogeneous graphs, with energy depositions on each detector plane forming nodes on planar subgraphs. The network utilizes a multihead attention message-passing mechanism to perform background filtering and semantic labeling on these graph nodes, identifying those associated with the primary physics interaction with 98.0% efficiency and labeling them according to particle type with 94.9% efficiency. The network operates directly on detector observables across multiple two-dimensional representations but utilizes a three-dimensional-context-aware mechanism to encourage consistency between these representations. Model inference takes 0.12 s / event on a CPU and 0.005 s / event batched on a GPU. This architecture is designed to be a general-purpose solution for particle reconstruction in neutrino physics, with the potential for deployment across a broad range of detector technologies, and offers a core convolution engine that can be leveraged for a variety of tasks beyond the two described in this paper.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2023

Transforming microseismic clouds into near real-time visualization of the growing hydraulic fracture

SUMMARY Microseismic observations during unconventional reservoir stimulation are typically seen as a proxy for clusters of hydraulic fractures and the extent of the stimulated reservoir. Such straightforward interpretation is often misleading and fails to provide a physically reasonable image of the fracturing process. This paper demonstrates the application of a physics-based machine learning algorithm which enables a rapid and accurate fracture mapping from the microseismic data. Our training and validation data set relies on a history-matched geomechanical modelling workflow implemented in GEOS software for the Hydraulic Fracturing Test Site 1 (HFTS-1) project. For this study we augmented the simulated fracture growth through geostatistical modelling of induced seismicity, so that the synthetic microseismic catalogue matches the main statistical properties of the field observations. We formulated the problem of mapping the actual fracture in the clutter of events to parallel common video segmentation workflows: several past video frames (microseismic density snapshots) are passed through a deep convolutional network to classify whether a given voxel is associated with a fracture or intact rock. We found that for accurate fracture mapping, the network’s input and architecture must be augmented to incorporate the fluid injection parameters (pressure, rate, concentration of proppant, and location of the perforation within the cluster). The error rate for the network reached as little as 10 per cent of the fracture area, while a conventional microseismic interpretation approach yielded ∼300 per cent. Our approach also yields must faster predictions than conventional methods (minutes instead of weeks), and could enable engineers to make rapid decisions regarding engineering parameters (pumping rate, viscosity) in real time during stimulation.

58 GEOSCIENCES↗