Search NASA⌕ Search

SEARCH · Search NASA

Results for “DL”

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 217 records · Page 12
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Toward a fully coherent tender and hard X-ray free-electron laser via cascaded EEHG in fourth-generation synchrotron light sources

Free-electron-laser-based beamlines utilize fully coherent laser pulses with extremely narrow bandwidth allowing direct use of X-rays without monochromators. This could be very beneficial for all users of current and future fourth-generation diffraction-limited synchrotron light sources (DL-SLSs) who need narrowband full-coherence high-brightness X-ray pulses. Based on our previous finding, i.e. that separating the two stages of echo-enabled harmonic generation (EEHG) with a few extra bending-magnet sections provides an effective way to increase the momentum compaction of chicane 1, one can simultaneously achieve adequate prebunching at extremely high harmonics as well as keep the energy modulation to the ideal minimum. This could open the door for cascaded EEHG, toward fully coherent tender and hard X-ray wavelengths. Built on our compact design of a twin-pulse seeding electron beam with an adjustable delay and timing jitter at the level of a few femtoseconds, a cascaded EEHG can be implemented, which includes two EEHG beamlines, where the radiation pulse generated by the first beamline with harmonic $h$ 1 could be used as the input seed laser pulse to the second beamline with harmonic $h$ 2 . Hence, the second radiator could potentially reach very high harmonics [$h$ = $h$ 1 (20)$h$ 2 (25–100)] from 500 to 2000, corresponding to tender and hard X-ray wavelengths. It is demonstrated that the cascaded EEHG scheme is compatible with almost any current or planned fourth-generation DL-SLS, with significant benefits for space-limited storage rings in particular. The main advantage is that this scheme requires almost no change of the storage-ring lattice and is fully compatible with other beamlines. Current proposals for rings with much longer straight sections would add self-amplified spontaneous emission as another viable option for storage-ring-based free-electron lasers.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Counter Data Paucity through Adversarial Invariance Encoding: A Case Study on Modeling Battery Thermal Runaway

Lithium-ion batteries, widely used for their durability and high energy storage, face the risk of internal short circuits leading to catastrophic thermal runaway events. These events, triggered by external stimuli like mechanical loads, pose safety concerns in applications such as electric vehicles. Detecting and understanding thermal runaway events is crucial, but physics-driven models struggle to explain the non-linear evolution of battery temperature during these events, considering factors like material composition and state-of-charge. Due to the rarity of these events and the cost of data collection, we propose a deep learning (DL) model to predict battery temperature responses during thermal runaway. The challenge lies in the scarcity of data, making traditional DL models prone to overfitting and learning low-quality representations of the complex process.Our approach introduces a novel few-shot architecture that incorporates an adversarially governed invariant encoding process. This architecture aims to distill "invariant" relationships by addressing distributional shifts in data across various battery properties, facilitating the detection of thermal runaway events. Specifically, our results demonstrate that deep learning models conditioned on these "invariant" representations outperform state-of-the-art baselines, achieving a remarkable 96.8% performance improvement in terms of the popular metric MAPE. This framework presents a promising direction for enhancing battery safety modeling, particularly in the context of rare and complex events like thermal runaway. Our code and code and dataset used for the paper are public1.

Tabassum, Anika [ORNL] (ORCID:0000000254600955)↗
DOE OSTI2022

Towards Low-Overhead Resilience for Data Parallel Deep Learning

Data parallel techniques have been widely adopted both in academia and industry as a tool to enable scalable training of deep learning models. At scale, DL training jobs can fail due to software or hardware bugs, may need to be preempted or terminated due to unexpected events, or may perform suboptimally because they were misconfigured. Under such circumstances, there is a need to recover and/or reconfigure data-parallel DL training jobs on-the-fly, while minimizing the impact on the accuracy of the DNN model and the runtime overhead. In this regard, state-of-art techniques adopted by the HPC community mostly rely on checkpoint-restart, which inevitably leads to loss of progress, thus increasing the runtime overhead. In this paper we explore alternative techniques that exploit the properties of modern deep learning frameworks (overlapping of gradient averaging and weight updates with local gradient computations through pipeline parallelism) to reduce the overhead of resilience/elasticity. To this end we introduce a failure simulation framework and two resilience strategies (immediate mini-batch rollback and lossy forward recovery), which we study compared with checkpoint-restart approaches in a variety of settings in order to understand the trade-offs between the accuracy loss of the DNN model and the runtime overhead.

data-parallel training↗
DOE OSTI2024

Distributed-Memory Sparse Deep Neural Network Inference Using Global Arrays

Partitioned Global Address Space (PGAS) models exhibit tremendous promise in developing efficient and productive distributed-memory parallel applications. They have been used extensively in scientific computations due to conveniently offering a ``shared-memory''-like model and convenient interfaces that separate communication with synchronization. Traditionally, PGAS communication models have been applied to dense/contiguously distributed data, but most modern applications depict varied levels of sparsity. Existing PGAS models require certain adaptations to support distributed sparse computations, since associated computations often require matrix arithmetic, in addition to data movement. The Global Arrays toolkit from Pacific Northwest National Laboratory (PNNL) is one of the earliest PGAS models to combine one-sided data communication and distributed matrix operations and is still used in the popular NWChem quantum chemistry suite. Recently, we have expanded the Global Arrays toolkit to support common sparse operations, like sparse matrix-dense matrix multiplies (SpMM), sparse matrix-sparse matrix multiplication (SpGEMM) and Sampled Dense-Dense Matrix Multiplication (SDDMM). As it turns out, these operations are the bedrock of sparse Deep Learning (DL); sparse deep neural networks and Graph Neural Networks (GNNs) have gained increasing attention recently in achieving speedups on training and inference with reduced memory footprints. Unlike scientific applications in High Performance Computing (HPC), modern (distributed-memory capable) DL toolkits often rely on non-standardized and closed-source vendor software optimizations, creating challenges in software-hardware co-design at scale. Our goal is to support a variety of distributed-memory sparse matrix operations and helper functions in the newly created Sparse Global Arrays (SGA), such that it is possible to build portable and productive Machine Learning scenarios for algorithm/software and hardware codesign purposes. Contemporary data-parallel schemes for training/inference are undergoing a major overhaul since model replication limits scalability and causes resource inefficiencies. As such, we have adopted tensor parallelism in decomposing the model and inputs, to mitigate memory issues. Current implementation is built on top of MPI and uses CPUs to maximize the portability across the platforms.

Distributed computing, machine learning↗
DOE OSTI2023

Uncertainty Aware Deep Learning for Fault Prediction Using Multivariate Time Series Signals

The superconducting radio-frequency cavities are a crucial component of the Continuous Electron Beam Accelerator Facility (CEBAF) at Jefferson Lab. When a cavity faults, beam delivery to experimental end users is disrupted. Prediction of cavity faults prior to onset is essential to reduce operation and maintenance costs. In this work, a parallel long short-term memory (LSTM)-convolution neural network (CNN)-based deep learning (DL) model is proposed to predict impending faults using pre-fault signals. Further, we introduce an uncertainty quantification approach using Monte Carlo dropout with the LSTM-CNN model to ascertain confidence in the prediction. The model was tested using multivariate time series signals from stable cavity operations and before faults. Initial results show that on the test dataset, the model can identify impending faults before their onset with an average 10-fold cross validation accuracy of 97.39% and a standard deviation of 0.12% using a 100-ms time window. It is also observed that the model performs better as the prediction time moves closer to the fault onset. For additional context, we compare the performance of the model with three machine-learning-based (ML) fault prediction models. Our proposed parallel LSTM-CNN-based DL method shows better performance than the ML-based methods.

Rahman, Md Monibor↗
DOE OSTI2026

Dynamic Validation of CNN-Based Surrogate Models for Inverter-Based Resources in Open-Source Solvers

Traditionally, distribution system planning has focused on steady-state analyses, with limited consideration of dynamic behavior. However, as large or medium-scale inverter-based resources (IBRs), particularly grid-following (GFL) inverters in commercial or industry buildings, become more prevalent, understanding their dynamic impact is essential for grid planning and operation. This article presents an innovative deep-learning (DL)-approach using convolutional neural networks technique to model the GFL inverters. Developed from real grid-tied commercial IBR transient data, these dynamic DL models overcome proprietary constraints by requiring minimal knowledge of internal converter physics while maintaining high accuracy and flexibility. To demonstrate their applicability, the models were incorporated into GridLAB-D, an open-source, three-phase distribution analysis tool. This integration enables dynamic simulations of large-scale distribution networks with high IBR penetration stability analysis. Rigorous testing and validation, aligned with industry standards, confirmed the reliability and efficiency of this approach, paving the way for enhanced planning and operational assessments of modern power systems.

Deep-learning↗
DOE OSTI2021

Unbundling Smart Meter Services Through Spatio-Temporal Decomposition Agents in DER-rich Environment

Smart meters and the advanced metering infrastructure (AMI) facilitate distribution system operators (DSOs) to gather information on energy consumption at the customer level. With the increasing penetration of building-level intermittent distributed energy resources (DERs) behind the meter, DER information is not available to DSOs. At the same time, smart meter enables users to participate in grid, with real-time information. Information for behind the meter is needed by user to coordinate building level assets for maximum benefits. The concept of unbundled smart meter (USM) needs agents to decompose smart meter measurements to provide service to DSO as well as customers. In this paper, we propose a Spatio-Temporal Decomposition Agent (STDA) for USM based on Artificial Intelligence (AI). STDA can help users optimize their energy usage, help DSO to utilize building assets for the grid operation. The energy usage strategy developed by STDA is suitable for different users, and can be customized by deep learning (DL) models according to the different energy consumption habits of each user. The power prediction performance results of various DL models and evaluation using a set of data from a Hawaii utility is presented. Furthermore, STDA integration with Home Energy management Systems (HEMS) to manage resources is presented and validated. STDA pre-processes the measurements before model training, and provides the spatio-temporal decomposed forecasting.

42 ENGINEERING↗
DOE OSTI2022

Frequency Domain Detection and Precoding for Massive MIMO With Single Carrier Modulation

Single carrier modulation (SCM) schemes are attractive for uplink (UL) transmissions due to improved power efficiency at the user equipment (UE) transmitter compared with multi-carrier modulation schemes. In a massive MIMO scenario with SCM, the UL detection must mitigate the effects of inter-symbol interference and multiuser interference. This processing is effectively performed in the frequency domain (FD) using a minimum mean squared error (MMSE) detector when the transmission is framed with a cyclic prefix. This paper presents a MMSEbased detector that is compatible with distributed processing in a time-division duplex (TDD) system. The matrix inverses computed for the UL detection are then reused to perform multi-user precoding for the downlink (DL). We find that this scheme yields a tremendous savings in computational complexity compared to commonly used zero-forcing (ZF) precoding without sacrificing any performance. Since MMSE processing introduces a bias to the estimates, we then derive the scalar coefficients necessary to cancel the MMSE bias. The impact of channel estimation errors are analyzed for both the UL and DL cases in conjunction with a power-efficient approach to SCM channel estimation. Moreover, extensive simulations are performed to confirm our theoretical findings.

42 ENGINEERING↗
DOE OSTI2023

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↗
DOE OSTI2021

Diagnosing autonomous vehicle driving criteria with an adversarial evolutionary algorithm

We repurposed an adversarial evolutionary algorithm, Gremlin, from finding driving scenarios where a model of an autonomous vehicle drove poorly to troubleshooting driving quality evaluation criteria. We evaluated the driving performance of a "perfect driver" robot in a virtual town environment using the same fitness criteria intended for a deep learner (DL) trained driver. We found that the fitness evaluation criteria poorly handled turns, and used Gremlin to iteratively improve that criteria. We were confident that the same criteria could then be applied to the DL-based models as originally intended, and that this approach could be used as a general means of troubleshooting autonomous vehicle driving criteria.

Coletti, Mark↗
DOE OSTI2020

Cryptococcus neoformans Secretes Small Molecules That Inhibit IL-1 β Inflammasome-Dependent Secretion

Cryptococcus neoformans is an encapsulated yeast that causes disease mainly in immunosuppressed hosts. It is considered a facultative intracellular pathogen because of its capacity to survive and replicate inside phagocytes, especially macrophages. This ability is heavily dependent on various virulence factors, particularly the glucuronoxylomannan (GXM) component of the polysaccharide capsule. Inflammasome activation in phagocytes is usually protective against fungal infections, including cryptococcosis. Nevertheless, recognition of C. neoformans by inflammasome receptors requires specific changes in morphology or the opsonization of the yeast, impairing proper inflammasome function. In this context, we analyzed the impact of molecules secreted by C. neoformans B3501 strain and its acapsular mutant Δcap67 in inflammasome activation in an in vitro model. Our results showed that conditioned media derived from B3501 was capable of inhibiting inflammasome-dependent events (i.e., IL-1β secretion and LDH release via pyroptosis) more strongly than conditioned media from Δcap67, regardless of GXM presence. We also demonstrated that macrophages treated with conditioned media were less responsive against infection with the virulent strain H99, exhibiting lower rates of phagocytosis, increased fungal burdens, and enhanced vomocytosis. Moreover, we showed that the aromatic metabolite DL-Indole-3-lactic acid (ILA) and DL-p-Hydroxyphenyllactic acid (HPLA) were present in B3501’s conditioned media and that ILA alone or with HPLA is involved in the regulation of inflammasome activation by C. neoformans. These results were confirmed by in vivo experiments, where exposure to conditioned media led to higher fungal burdens in Acanthamoeba castellanii culture as well as in higher fungal loads in the lungs of infected mice. Overall, the results presented show that conditioned media from a wild-type strain can inhibit a vital recognition pathway and subsequent fungicidal functions of macrophages, contributing to fungal survival in vitro and in vivo and suggesting that secretion of aromatic metabolites, such as ILA, during cryptococcal infections fundamentally impacts pathogenesis.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2022

LivermorE Al Projector for Computed Tomography Tasks

With recent computed tomography (CT) efforts using Artificial Intelligence (AI) and Deep Learning (DL)techniques, there is a strong need for differentiable forward projection models that can be integrated into existing DL frameworks. We developed a pytorch-based package library providing differentiable forward and back projection functions and classes to facilitate forward and back propagation of CT operations in the training procedure. This forward projectors support three CT projection geometries: cone, parallel and modular beams. This package can be used with both CPU and GPU with CUDA.

Kim, Hyojin↗
DOE OSTI2022

Deep Neural Network Algorithm for CMC Microstructure Characterization and Variability Quantification

Microstructure characterization and variability quantification are crucial for understanding ceramic matrix composites (CMCs) mechanical behavior and deformation mechanisms across length scales. Traditionally, analyses of the micrographs obtained from microscopy are labor-intensive. However, with the vast improvement in computer vision (CV) and deep learning (DL), an automated algorithm can be designed to extract essential microstructure variability from micrographs which can then be used to construct a statistically representative volume element (SRVE). The DL-based algorithm spans the taxonomy of microstructure analyses, including semantic segmentation of microstructure constituents, secondary phases, matrix/fiber interface, and defects, and quantifying the microstructure variability in terms of probability distributions. In this work, C/SiNC and SiC/SiNC CMCs microstructures are semantically segmented through a deep convolutional neural network, followed by variability quantification through the implementation of a fully connected regression layer, hence forming a deep regression network. The deep regression network operates in a feedforward regime, in which the neuron output signal traverses through the network in a unidirectional manner. The weight tensor associated with each layer is updated through a backpropagation stochastic gradient descent approach. The input gray-scale image obtained through in-house scanning electron microscope and confocal microscope micrographs is augmented through affine transformations to increase the training set size, which is then processed through four strided convolutional layers. This compresses the image resolution by half at each layer while increasing the image depth by applying different filters (image encoding). The class activation maps (CAMs) corresponding to the applied filters highlight the key architectural features and assist with the semantic segmentation of the microstructure.

Hamza, Mohamed H.↗
DOE OSTI2022

Characterizing warm atmospheric boundary layer over land by combining Raman and Doppler lidar measurements

PBL plays a critical role in the atmosphere by transferring heat, moisture, and momentum. The warm PBL has a distinct diurnal cycle including daytime convective mixing layer (ML) and nighttime residual layer developments. Thus, for PBL characterization and process study, simultaneous determinations of PBL height (PBLH) and ML height (MLH) are necessary. Here, new approaches are developed to provide reliable PBLH and MLH to characterize warm PBL evolution. The approaches use Raman lidar (RL) water vapor mixing ratio (WVMR) and Doppler lidar (DL) vertical velocity measurements at the Southern Great Plains (SGP) atmospheric observatory, which was established by the Atmospheric Radiation Measurement (ARM) user facility. Compared with widely used lidar aerosol measurements for PBLH, WVMR is a better trace for PBL vertical mixing. For PBLH, the approach classifies PBL water vapor structures into a few general patterns, then uses a slope method and dynamic threshold method to determine PBLH. For MLH, wavelet analysis is used to re-construct 2-D variance from DL vertical wind velocity measurements according to the turbulence eddy size to minimize the impacts of gravity wave and eddy size on variance calculations; then, a dynamic threshold method is used to determine MLH. Remotely-sensed PBLHs and MLHs are compared with radiosonde measurements based on the Richardson number method. Good agreements between them confirm that the proposed new algorithms are reliable for PBLH and MLH characterization. The algorithms are applied to warm seasons’ RL and ML measurements at the SGP site for five years to study warm season PBL structure and processes. The weekly composited diurnal evolutions of PBLHs and MLHs in warm climate were provided to illustrate diurnal and seasonal PBL evolutions. This reliable PBLH and MLH dataset will be valuable for PBL process study, model evolution, and PBL parameterization improvement.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2020

Limited-view Cone Beam CT reconstruction using 3D Patch-based Supervised and Adversarial Learning [Slides]

We present a novel machine learning CNN architecture that can learn from limited data combined appropriately with physics and statistical priors (e.g., forward models and noise models). To address the limited availability of training data we adopt a 3D patch-based approach for our models. Patch-based learning is central to several image reconstruction methods and demands fewer training data than DL approaches, as a single data volume can be broken into several millions of overlapping 3D sub-volumes or patches. This creates a very large number of training sub-volumes from a limited number of overall image volumes. A 3D Generative Adversarial Networks (GAN) is then trained to remove artifacts at the sub-volume level. The combination of a sub-volume-based approach with DL allows us to exploit the richness of the latter in extracting and representing image features, while avoiding risks associated with overfitting due to limited training data.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2020

Analysis of Respirator Cartridge Performance Testing on Hanford Tank A-101

Washington River Protection Solutions (WRPS) conducted tests on two types of chemical cartridges for use in air purifying respirators (APR) to determine the period of time that the cartridges would provide adequate performance1 to protect workers when exposed to a mixture of Chemicals of Potential Concern (COPCs) from vapors emanating from the headspace of tank A-101 on the Hanford Site. The Occupational Safety and Health Administration (OSHA) identifies cartridge testing as a valid approach for establishing a cartridge change schedules.[3] Testing is commonly applied in situations where mixtures of COPCs exist, and where other approaches, such as manufacturer recommendations and modeling, are less reliable. The tests were designed and conducted to ensure measurement and/or control of the key variables OSHA identified as important to estimate cartridge service life, including temperature, humidity, COPC concentration, breathing rate, and cartridge adsorption capacity. Testing was conducted over a period from July 22-24, 2016, using headspace vapors from Hanford tank A-101 under static conditions2 fed to a respirator cartridge test stand developed by WRPS in collaboration with HiLine Engineering (Richland, Washington). Multipurpose respirator cartridges, SCOTT 7422-SD1 and 7422-SC1 (SCOTT Safety, Monroe, North Carolina) were assessed on separate days with A-101 headspace vapors. Sample media (sorbent tubes) were used to collect samples of the vapor stream entering and exiting the respirator cartridge, and were subsequently analyzed for COPC concentrations. Pacific Northwest National Laboratory was tasked with conducting an independent analysis of the analytical results and making recommendations based on the results for respiratory cartridge performance and service life. Key conclusions from the assessment of the 59 COPCs in this study are described below: • Based on d measured cartridge inlet vapor concentrations from tank A-101 headspace, both ammonia and N-Nitrosodimethylamine (NDMA) exceeded their Occupational Exposure Limits (OEL).3 These measurements were generally consistent with maximum A-101 headspace measurements previously obtained for these two compounds. • Ammonia had respirator cartridge outlet concentrations that exceeded 10% of the OEL for both cartridges tested, indicating breakthrough for each. For the SCOTT 7422-SD1 cartridge, ammonia appeared to breakthrough above 10% of its OEL after 2 hours. For the SCOTT 7422-SC1 cartridge, ammonia breakthrough appeared to occur after 8 hours. 1 “Adequate performance” refers to the breakthrough criteria used in this analysis—that is, sustained cartridge outlet concentrations above 10% of the OEL. For some COPCs, an alternate threshold has been applied when necessary due to higher detection limit (DL)/reporting limit (RL) values for specific compounds. Ultimately, Industrial Hygiene professionals will use these results along with specific hazard assessments to determine service life and cartridge use that provides the necessary performance. 2 These tests were conducted under static conditions absent waste-disturbing activities in the subject tank or tank farm. 3 Occupational Exposure Limits accepted for Hanford Tank Farm use are based on OELs established by a U.S. governmental agency or national professional organization (e.g., OSHA, National Institute for Occupational Safety and Health, American Conference of Governmental Industrial Hygienists), or if no U.S. OEL exists, standard toxicological practices are applied to develop OELs using non-U.S. exposure limits, other established OELs for chemical surrogates when available, or other standard procedures. The OEL for NDMA was established in 2005 based on the MAK (Maximale Arbeitsplatzkonzentration) Commission standard adopted in Europe. iv • Despite respirator inlet measurements for NDMA that were in excess of its OEL, all corresponding outlet measurements, from both respirator cartridges, were below analytical DL 1 for the

61 RADIATION PROTECTION AND DOSIMETRY↗
DOE OSTI2022

Deep learning-based spatio-temporal estimate of greenhouse gas emissions using satellite data

Accurate estimation of greenhouse gases (GHGs) emissions is very important for developing mitigation strategies to climate change by controlling and reducing GHG emissions. This project aims to develop multiple deep learning approaches to estimate anthropogenic greenhouse gas emissions using multiple types of satellite data. NO2 concentration is chosen as an example of GHGs to evaluate the proposed approach. Two sentinel satellites (sentinel-2 and sentinel-5P) provide multiscale observations of GHGs from 10-60m resolution (sentinel-2) to ~kilometer scale resolution (sentinel-5P). Among multiple deep learning (DL) architectures evaluated, two best DL models demonstrate that key features of spatio-temporal satellite data and additional information (e.g., observation times and/or coordinates of ground stations) can be extracted using convolutional neural networks and feed forward neural networks, respectively. In particular, irregular time series data from different NO 2 observation stations limit the flexibility of long short-term memory architecture, requiring zero-padding to fill in missing data. However, deep neural operator (DNO) architecture can stack time-series data as input, providing the flexibility of input structure without zero-padding. As a result, the DNO outperformed other deep learning architectures to account for time-varying features. Overall, temporal patterns with smooth seasonal variations were predicted very well, while frequent fluctuation patterns were not predicted well. In addition, uncertainty quantification using conformal inference method is performed to account for prediction ranges. Overall, this research will lead to a new groundwork for estimating greenhouse gas concentrations using multiple satellite data to enhance our capability of tracking the cause of climate change and developing mitigation strategies.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

2022 AI Testbed Expeditions Report

By exploiting the coherent properties of a light source, coherent diffraction imaging (CDI) is able to obtain the sample image at a nanoscale resolution using the measured diffraction pattern. Bragg Coherent Diffraction Imaging (BCDI) has become valuable for recovering the displacement and strain field of crystals, providing a valuable tool in material science and solid-state physics. X-ray ptychography is another emerging CDI technique that can produce a high-resolution image of the extended sample and has become popular in many research areas (e.g., materials science, biology, electronics, and optics characterization). CDI including BCDI and ptychography has become an established technique in Synchrotron Facilities including the Advanced Photon Source (APS) and will greatly benefit from the 100x coherent flux increase of the upcoming APS Upgrade (APSU). The current image formation process in CDI employs iterative phase retrieval algorithms, which is a time-consuming and computationally expensive process. Especially after APSU, the traditional iterative methods will not be able to match the experimental data acquisition speed. We employ deep learning (DL) approach to replace the iterative approaches, therefore allowing hundreds of times faster recovery of the object. We developed AutoPhaseNN, a DL-based approach which learns to solve the inverse problem without labeled data. Taking 3D BCDI as a representative technique, AutoPhaseNN has been demonstrated to be one hundred times faster than traditional iterative phase retrieval methods while providing comparable image quality. The current network is trained with 64 x 64 x 64 data size, to achieve higher resolution imaging, we will need to scale the network to input and train/infer 3D arrays of size 256 x 256 x 256 (today) and of size 2560x2560x2560 (APSU). However, the scalability of the network is restricted due to the memory-intensive training process. To perform the training for a 256 x 256 x 256 data size, the required memory exceeds the capacity of the current machine. In this project, we explore using Sambanova system to train the network for the direct data inversion for CDI.

36 MATERIALS SCIENCE↗