Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Deep 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 1,189 records · Page 66

Bayesian Inference with Latent Hamiltonian Neural Networks (L-HNNs)

When sampling for Bayesian inference, one popular approach is to use Hamiltonian Monte Carlo (HMC) and the No-U-Turn Sampler (NUTS). However, HMC and NUTS can require numerous numerical gradients of the target density and can prove slow in practice. We propose Hamiltonian neural networks (HNNs) with HMC and NUTS for solving Bayesian inference problems [1, 2]. Once trained, HNNs do not require gradients of the target density while sampling. Moreover, they satisfy important properties such as perfect time reversibility and Hamiltonian conservation, making them well suited for use within HMC and NUTS because stationarity can be shown. We also propose an HNN extension called latent HNNs (L-HNNs), which predict latent variable outputs. Compared to HNNs, L-HNNs offer improved expressivity and a reduction in integration errors. Finally, we propose employing L-HNNs in NUTS with an online error monitoring scheme to prevent degeneracy of the sampling in regions of low probability density. We demonstrate L-HNNs in NUTS with online error monitoring by using several example cases involving complex, heavy-tailed, and high local curvature probability densities. Overall, L-HNNs in NUTS with online error monitoring satisfactorily inferred these probability densities. Compared to traditional NUTS, L-HNNs in NUTS with online error monitoring improved the effective sample size (ESS) per gradient by an order of magnitude.

97 MATHEMATICS AND COMPUTING↗

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↗

Scaling Deep Learning workloads: NVIDIA DGX-1/Pascal and Intel Knights Landing

Deep Learning (DL) algorithms have become ubiquitous in data analytics. As a result, major computing vendors—including NVIDIA, Intel, AMD, and IBM—have architectural road maps influenced by DL workloads. Furthermore, several vendors have recently advertised new computing products as accelerating large DL workloads. Unfortunately, it is difficult for data scientists to quantify the potential of these different products. Here, this article provides a performance and power analysis of important DL workloads on two major parallel architectures: NVIDIA DGX-1 (eight Pascal P100 GPUs interconnected with NVLink) and Intel Knights Landing (KNL) CPUs interconnected with Intel Omni-Path or Cray Aries. Our evaluation consists of a cross section of convolutional neural net workloads: CifarNet, AlexNet, GoogLeNet, and ResNet50 topologies using the Cifar10 and ImageNet datasets. The workloads are vendor-optimized for each architecture. We use sequentially equivalent implementations to maintain iso-accuracy between parallel and sequential DL models. Our analysis indicates that although GPUs provide the highest overall performance, the gap can close for some convolutional networks; and the KNL can be competitive in performance/watt. We find that NVLink facilitates scaling efficiency on GPUs. However, its importance is heavily dependent on neural network architecture. Furthermore, for weak-scaling—sometimes encouraged by restricted GPU memory—NVLink is less important.

97 MATHEMATICS AND COMPUTING↗

PPINN: Parareal physics-informed neural network for time-dependent PDEs

Physics-informed neural networks (PINNs) encode physical conservation laws and prior physical knowledge into the neural networks, ensuring the correct physics is represented accurately while alleviating the need for supervised learning to a great degree. While effective for relatively short-term time integration, when long time integration of the time-dependent PDEs is sought, the time–space domain may become arbitrarily large and hence training of the neural network may become prohibitively expensive. To this end, we develop a parareal physics-informed neural network (PPINN), hence decomposing a long-time problem into many independent short-time problems supervised by an inexpensive/fast coarse-grained (CG) solver. In particular, the serial CG solver is designed to provide approximate predictions of the solution at discrete times, while initiate many fine PINNs simultaneously to correct the solution iteratively. There is a two-fold benefit from training PINNs with small-data sets rather than working on a large-data set directly, i.e., training of individual PINNs with small-data is much faster, while training the fine PINNs can be readily parallelized. Consequently, compared to the original PINN approach, the proposed PPINN approach may achieve a significant speed-up for long-time integration of PDEs, assuming that the CG solver is fast and can provide reasonable predictions of the solution, hence aiding the PPINN solution to converge in just a few iterations. To investigate the PPINN performance on solving time-dependent PDEs, we first apply the PPINN to solve the Burgers equation, and subsequently we apply the PPINN to solve a two-dimensional nonlinear diffusion–reaction equation. Furthermore, our results demonstrate that PPINNs converge in a few iterations with significant speed-ups proportional to the number of time-subdomains employed.

42 ENGINEERING↗

Deep learning with mixup augmentation for improved pore detection during additive manufacturing

In additive manufacturing (AM), process defects such as keyhole pores are difficult to anticipate, affecting the quality and integrity of the AM-produced materials. Hence, considerable efforts have aimed to predict these process defects by training machine learning (ML) models using passive measurements such as acoustic emissions. This work considered a dataset in which keyhole pores of a laser powder bed fusion (LPBF) experiment were identified using X-ray radiography and then registered both in space and time to acoustic measurements recorded during the LPBF experiment. Due to AM’s intrinsic process controls, where a pore-forming event is relatively rare, the acoustic datasets collected during monitoring include more non-pores than pores. In other words, the dataset for ML model development is imbalanced. Moreover, this imbalanced and sparse data phenomenon remains ubiquitous across many AM monitoring schemes since training data is nontrivial to collect. Hence, we propose a machine learning approach to improve this dataset imbalance and enhance the prediction accuracy of pore-labeled data. Specifically, we investigate how data augmentation helps predict pores and non-pores better. This imbalance is improved using recent advances in data augmentation called Mixup, a weak-supervised learning method. Convolutional neural networks (CNNs) are trained on original and augmented datasets, and an appreciable increase in performance is reported when testing on five different experimental trials. When ML models are trained on original and augmented datasets, they achieve an accuracy of 95% and 99% on test datasets, respectively. We also provide information on how dataset size affects model performance. Lastly, we investigate the optimal Mixup parameters for augmentation in the context of CNN performance.

36 MATERIALS SCIENCE↗

Deep learning with mixup augmentation for improved pore detection during additive manufacturing

In additive manufacturing (AM), process defects such as keyhole pores are difficult to anticipate, affecting the quality and integrity of the AM-produced materials. Hence, considerable efforts have aimed to predict these process defects by training machine learning (ML) models using passive measurements such as acoustic emissions. This work considered a dataset in which keyhole pores of a laser powder bed fusion (LPBF) experiment were identified using X-ray radiography and then registered both in space and time to acoustic measurements recorded during the LPBF experiment. Due to AM’s intrinsic process controls, where a pore-forming event is relatively rare, the acoustic datasets collected during monitoring include more non-pores than pores. In other words, the dataset for ML model development is imbalanced. Moreover, this imbalanced and sparse data phenomenon remains ubiquitous across many AM monitoring schemes since training data is nontrivial to collect. Hence, we propose a machine learning approach to improve this dataset imbalance and enhance the prediction accuracy of pore-labeled data. Specifically, we investigate how data augmentation helps predict pores and non-pores better. This imbalance is improved using recent advances in data augmentation called Mixup, a weak-supervised learning method. Convolutional neural networks (CNNs) are trained on original and augmented datasets, and an appreciable increase in performance is reported when testing on five different experimental trials. When ML models are trained on original and augmented datasets, they achieve an accuracy of 95% and 99% on test datasets, respectively. We also provide information on how dataset size affects model performance. Lastly, we investigate the optimal Mixup parameters for augmentation in the context of CNN performance.

42 ENGINEERING↗

Deep learning-based predictive models for laser direct drive at the Omega Laser Facility

The rich and complex physics of inertial confinement fusion provides a unique and challenging space for high-fidelity first-principles modeling. Consequently, simulation codes that are used to design experiments are computationally expensive and lack the predictive capability required for extensive parameter exploration in search of a high-performing design for laser direct drive. In this article, we present two deep-learning-based predictive models intended to address these difficulties. The first model (TL DNN) acts as a fast emulator of simulations as well as experiments at the Omega Laser Facility. This model is trained on a simulation database and subsequently calibrated on experimental data using transfer learning. To facilitate the development of this model, an autoencoder is developed to reduce the dimensionality of the input space by compressing the laser pulse input. The model predicts key experimental scalar observables of Omega experiments with high accuracy and minimal computational cost. This deep neural net enables rapid exploration of a high-dimensional input parameter space for an optimal implosion design. The second model (DNN SM+) aims to extend the statistical modeling work of Lees et al. [Phys. Rev. Lett. 127, 105001 (2021)], by increasing the complexity of the model space and allowing for coupling between degradation terms. Since the model capacity of DNN SM+ is higher than the model of Lees et al., DNN SM+ can potentially provide an improvement in predictive capability, and we use this model to provide insight into complicated degradation dependencies.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

DeepAstroUDA: semi-supervised universal domain adaptation for cross-survey galaxy morphology classification and anomaly detection

Abstract Artificial intelligence methods show great promise in increasing the quality and speed of work with large astronomical datasets, but the high complexity of these methods leads to the extraction of dataset-specific, non-robust features. Therefore, such methods do not generalize well across multiple datasets. We present a universal domain adaptation method, DeepAstroUDA , as an approach to overcome this challenge. This algorithm performs semi-supervised domain adaptation (DA) and can be applied to datasets with different data distributions and class overlaps. Non-overlapping classes can be present in any of the two datasets (the labeled source domain, or the unlabeled target domain), and the method can even be used in the presence of unknown classes. We apply our method to three examples of galaxy morphology classification tasks of different complexities (three-class and ten-class problems), with anomaly detection: (1) datasets created after different numbers of observing years from a single survey (Legacy Survey of Space and Time mock data of one and ten years of observations); (2) data from different surveys (Sloan Digital Sky Survey (SDSS) and DECaLS); and (3) data from observing fields with different depths within one survey (wide field and Stripe 82 deep field of SDSS). For the first time, we demonstrate the successful use of DA between very discrepant observational datasets. DeepAstroUDA is capable of bridging the gap between two astronomical surveys, increasing classification accuracy in both domains (up to 40 % on the unlabeled data), and making model performance consistent across datasets. Furthermore, our method also performs well as an anomaly detection algorithm and successfully clusters unknown class samples even in the unlabeled target dataset.

79 ASTRONOMY AND ASTROPHYSICS↗

Probing Electron Beam Induced Transformations on a Single-Defect Level via Automated Scanning Transmission Electron Microscopy

A robust approach for real-time analysis of the scanning transmission electron microscopy (STEM) data streams, based on ensemble learning and iterative training (ELIT) of deep convolutional neural networks, is implemented on an operational microscope, enabling the exploration of the dynamics of specific atomic configurations under electron beam irradiation via an automated experiment in STEM. Combined with beam control, this approach allows studying beam effects on selected atomic groups and chemical bonds in a fully automated mode. Here, in this study, we demonstrate atomically precise engineering of single vacancy lines in transition metal dichalcogenides and the creation and identification of topological defects in graphene. The ELIT-based approach facilitates direct on-the-fly analysis of the STEM data and engenders real-time feedback schemes for probing electron beam chemistry, atomic manipulation, and atom by atom assembly.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Efficient Bayesian inference with latent Hamiltonian neural networks in No-U-Turn Sampling

When sampling for Bayesian inference, one popular approach in the computational field is to use Hamiltonian Monte Carlo (HMC) and specifically the No-U-Turn Sampler (NUTS), which automatically decides the end time of the Hamiltonian trajectory. However, HMC and NUTS can require numerous numerical gradients of the target density and can prove slow in practice when relying on computationally expensive forward models. We propose Latent Hamiltonian neural networks (L-HNNs) with HMC and NUTS for solving Bayesian inference problems. Once trained, L-HNNs do not require numerical gradients of the target density during sampling, and hence numerous evaluations of the forward computational model. Moreover, L-HNNs satisfy important properties such as perfect time reversibility and Hamiltonian conservation, making them well-suited for use within HMC and NUTS because stationarity can be shown. We also propose the integration of L-HNNs in an online error monitoring scheme, in which numerical gradients of the target density are used for a few samples whenever the L-HNNs prediction errors are large. This online error monitor scheme prevents sample degeneracy in regions of low probability density and ensures robust uncertainty quantification. We demonstrate L-HNNs in NUTS with online error monitoring on several analytical examples involving complex, heavy-tailed, and high-local-curvature probability densities. We then demonstrate the applicability of L-HNNs in NUTS to two computational case studies, namely the Allen-Cahn stochastic partial differential equation and an elliptic partial differential equation with 25 and 50 inference parameters, respectively. Overall, the L-HNNs in NUTS with online error monitoring satisfactorily inferred these probability densities. In conclusion, compared to traditional NUTS, L-HNNs in NUTS with online error monitoring required 1–2 orders of magnitude fewer numerical gradients of the target density and improved the effective sample size (ESS) per gradient (which is a measure of both the sampling quality and the computational expense) by an order of magnitude.

97 MATHEMATICS AND COMPUTING↗

Effect of image resolution on automated classification of chest X-rays

Deep learning (DL) models have received much attention lately for their ability to achieve expert-level performance on the accurate automated analysis of chest X-rays (CXRs). Recently available public CXR datasets include high resolution images, but state-of-the-art models are trained on reduced size images due to limitations on graphics processing unit memory and training time. As computing hardware continues to advance, it has become feasible to train deep convolutional neural networks on high-resolution images without sacrificing detail by downscaling. This study examines the effect of increased resolution on CXR classification performance. We used the publicly available MIMIC-CXR-JPG dataset, comprising 377,110 high resolution CXR images for this study. We applied image downscaling from native resolution to 2048 × 2048 pixels, 1024 × 1024 pixels, 512 × 512 pixels, and 256 × 256 pixels and then we used the DenseNet121 and EfficientNet-B4 DL models to evaluate clinical task performance using these four downscaled image resolutions. We find that while some clinical findings are more reliably labeled using high resolutions, many other findings are actually labeled better using downscaled inputs. We qualitatively verify that tasks requiring a large receptive field are better suited to downscaled low resolution input images, by inspecting effective receptive fields and class activation maps of trained models. Lastly, we show that stacking an ensemble across resolutions outperforms each individual learner at all input resolutions while providing interpretable scale weights, indicating that diverse information is extracted across resolutions.

47 OTHER INSTRUMENTATION↗

Performance analysis and optimization for scalable deployment of deep learning models for country-scale settlement mapping on Titan supercomputer

Here, we present a scalable object detection workflow for detecting objects, such as settlements, from remotely sensed (RS) imagery. We have successfully deployed this workflow on Titan supercomputer and utilized it for the task of mapping human settlement at a country scale. The performance of various stages in the workflow was analyzed before making it operational. The workflow implemented various strategies to address issues such as suboptimal resource utilization and long-tail effects due to unbalanced image workload, data loss due to runtime failures, and maximum wall-time constraints imposed by Titan's job scheduling policy. A mean shift clustering–based static load balancing strategy was implemented, which partitions the image load such that each partition contained similar-sized images. Furthermore, a checkpoint-restart strategy was added in the workflow as a fault-tolerance mechanism to prevent the data losses due to unforeseen runtime failures. The performance of the above-mentioned strategies was observed in various scenarios, such as node failure, exceeding wall time, and successful completion. Using this workflow, we have examined an RS data set that has a spatial resolution of 0.31 m and is comprised of 685 675 km 2 of area of the Republic of Zambia in under six hours using 5426 nodes of the Titan supercomputer.

97 MATHEMATICS AND COMPUTING↗

Deep learning-driven super-resolution in Raman hyperspectral imaging: Efficient high-resolution reconstruction from low-resolution data

Deep learning (DL) has become an indispensable tool in hyperspectral data analysis, automatically extracting valuable features from complex, high-dimensional datasets. Super-resolution reconstruction, an essential aspect of hyperspectral data, involves enhancing spatial resolution, particularly relevant to low-resolution hyperspectral data. Yet, the pursuit of super-resolution in hyperspectral analysis is fraught with challenges, including acquiring ground truth high-resolution data for training, generalization, and scalability. The pressing issue of extended spectral acquisition times, notably for high-resolution scans, is a significant roadblock in hyperspectral imaging. Super-resolution methods offer a promising solution by providing higher spatial resolution data to expedite data collection and yield more efficient outcomes. This paper delves into a practical application of these concepts using Raman imaging, where spectral acquisition times can be prohibitively long. In this context, DL-based super-resolution models demonstrate their efficacy by predicting and reconstructing high-resolution Raman data from low-resolution input, eliminating the need for resource-intensive high-resolution scans. While previous work often relied on substantial high-resolution datasets, this study showcases the ability to achieve similar outcomes even with limited data, presenting a more practical and cost-effective approach. In conclusion, the results offer a glimpse into the transformative potential of this technology to streamline hyperspectral imaging applications by saving valuable time and resources through the successful generation of high-resolution data from low-resolution inputs.

42 ENGINEERING↗

Autoencoders for semivisible jet detection

The production of dark matter particles from confining dark sectors may lead to many novel experimental signatures. Depending on the details of the theory, dark quark production in proton-proton collisions could result in semivisible jets of particles: collimated sprays of dark hadrons of which only some are detectable by particle collider experiments. The experimental signature is characterised by the presence of reconstructed missing momentum collinear with the visible components of the jets. This complex topology is sensitive to detector inefficiencies and mis-reconstruction that generate artificial missing momentum. With this work, we propose a signal-agnostic strategy to reject ordinary jets and identify semivisible jets via anomaly detection techniques. A deep neural autoencoder network with jet substructure variables as input proves highly useful for analyzing anomalous jets. The study focuses on the semivisible jet signature; however, the technique can apply to any new physics model that predicts signatures with anomalous jets from non-SM particles.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

High-dimensional anomaly detection with radiative return in $\mathcal{e}$⁺$\mathcal{e}$⁻ collisions

Experiments at a future $\mathcal{e}$⁺$\mathcal{e}$⁻ collider will be able to search for new particles with masses below the nominal centre-of-mass energy by analyzing collisions with initial-state radiation (radiative return). We show that machine learning methods that use imperfect or missing training labels can achieve sensitivity to generic new particle production in radiative return events. In addition to presenting an application of the classification without labels (CWoLa) search method in $\mathcal{e}$⁺$\mathcal{e}$⁻ collisions, our study combines weak supervision with variable-dimensional information by deploying a deep sets neural network architecture. We have also investigated some of the experimental aspects of anomaly detection in radiative return events and discuss these in the context of future detector design.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Boosting background suppression in the NEXT experiment through Richardson-Lucy deconvolution

Next-generation neutrinoless double beta decay experiments aim for half-life sensitivities of ~ 10 27 yr, requiring suppressing backgrounds to < 1 count/tonne/yr. For this, any extra background rejection handle, beyond excellent energy resolution and the use of extremely radiopure materials, is of utmost importance. The NEXT experiment exploits differences in the spatial ionization patterns of double beta decay and single-electron events to discriminate signal from background. While the former display two Bragg peak dense ionization regions at the opposite ends of the track, the latter typically have only one such feature. Thus, comparing the energies at the track extremes provides an additional rejection tool. The unique combination of the topology-based background discrimination and excellent energy resolution (1% FWHM at the Q-value of the decay) is the distinguishing feature of NEXT. Previous studies demonstrated a topological background rejection factor of ~ 5 when reconstructing electron-positron pairs in the 208Tl 1.6 MeV double escape peak (with Compton events as background), recorded in the NEXT-White demonstrator at the Laboratorio Subterráneo de Canfranc, with 72% signal efficiency. This was recently improved through the use of a deep convolutional neural network to yield a background rejection factor of ~ 10 with 65% signal efficiency. Here, we present a new reconstruction method, based on the Richardson-Lucy deconvolution algorithm, which allows reversing the blurring induced by electron diffusion and electroluminescence light production in the NEXT TPC. The new method yields highly refined 3D images of reconstructed events, and, as a result, significantly improves the topological background discrimination. When applied to real-data 1.6 MeV e - e + pairs, it leads to a background rejection factor of 27 at 57% signal efficiency.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Search for a new $Z$' gauge boson in 4$μ$ events with the ATLAS experiment

This paper presents a search for a new Z' vector gauge boson with the ATLAS experiment at the Large Hadron Collider using pp collision data collected at $\sqrt{s}$ = 13 TeV, corresponding to an integrated luminosity of 139 fb -1 . The new gauge boson Z' is predicted by Lμ - Lτ models to address observed phenomena that cannot be explained by the Standard Model. The search examines the four-muon (4μ) final state, using a deep learning neural network classifier to separate the Z' signal from the Standard Model background events. The di-muon invariant masses in the 4μ events are used to extract the Z' resonance signature. No significant excess of events is observed over the predicted background. Upper limits at a 95% confidence level on the Z' production cross-section times the decay branching fraction of pp → Z'μμ → 4μ are set from 0.31 to 4.3 fb for the Z' mass ranging from 5 to 81 GeV. The corresponding common coupling strengths, g Z' , of the Z' boson to the second and third generation leptons above 0.003 – 0.2 have been excluded.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Automated segmentation of porous thermal spray material CT scans with predictive uncertainty estimation

Abstract Thermal sprayed metal coatings are used in many industrial applications, and characterizing the structure and performance of these materials is vital to understanding their behavior in the field. X-ray computed tomography (CT) enables volumetric, nondestructive imaging of these materials, but precise segmentation of this grayscale image data into discrete material phases is necessary to calculate quantities of interest related to material structure. In this work, we present a methodology to automate the CT segmentation process as well as quantify uncertainty in segmentations via deep learning. Neural networks (NNs) have been shown to excel at segmentation tasks; however, memory constraints, class imbalance, and lack of sufficient training data often prohibit their deployment in high resolution volumetric domains. Our 3D convolutional NN implementation mitigates these challenges and accurately segments full resolution CT scans of thermal sprayed materials with maps of uncertainty that conservatively bound the predicted geometry. These bounds are propagated through calculations of material properties such as porosity that may provide an understanding of anticipated behavior in the field.

Martinez, Carianne↗