Search NASA⌕ Search

SEARCH · Search NASA

Results for “data gaps”

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 649 records · Page 36
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

High Level Gap Analysis for Accident Tolerant and Advanced Fuels for Storage and Transportation

This initial gap analysis considers proposed accident tolerant fuel (ATF) options currently being irradiated in commercial reactors, since these are most likely for future batch implementation. Also, advanced fuel (AF) options that may be likely for use in advanced reactors are considered. The cladding technologies considered were chromium-coated zirconium-based alloys, FeCrAl, and both monolithic and matrix composite Silicide carbide (SiC). The fuel technologies considered were chromium-doped uranium dioxide fuel, uranium alloys, uranium nitride, and uranium silicide. Numerous national labs, industry, and countries are performing significant testing and modeling on these proposed technologies to establish performance, but at this time none of the prototypes being irradiated have achieved end-of-life (EOL) burnup. There are some testing results after one burnup cycle to verify in-reactor performance, but little data beyond that. As the ATF prototypes acquire more burnup, data will be produced that is relevant to storage and transportation. The DOE:NE Spent Fuel and Waste Science and Technology (SWFST) Storage and Transportation (ST) Control Account will evaluate the performance data as it becomes available for application to the identified gaps for ST.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2024

Parameter extraction approaches for compact modeling of thermoelectric modules

Thermoelectric (TE) cooling has experienced rapid advancements with the foundational understanding of TE materials. TE modules, compact and lightweight devices, have become the prevalent approach for implementing TE technologies. Accurately quantifying TE physical parameters (Seebeck coefficient α, thermal conductivity κ, and thermal resistance ρ) is challenging due to the dynamic temperature changes in operation. Furthermore, extracting lumped property parameters is crucial for designing energy systems using TE modules. Existing research has several limitations, such as lack of comparative analysis between prevalent formulae, reliance on potentially inaccurate vendor-supplied data, disregard for fundamental assumptions, and absence of empirical measurements. Further, this study addresses these gaps by conducting TE material characterization, comparing three existing formulae using vendor datasheets, designing a laboratory test facility for model validation and refinement, and outlining a structured data extraction procedure. The study's novelty lies in multiple key contributions: (1) a detailed comparative analysis of existing formulae for extracting TE property parameter; (2) executing experimental work in a laboratory setting to validate the model and elucidate its limitations; (3) highlighting potential risks; (4) clarifying possible assumptions from both material and engineering perspectives; and (5) considering temperature differential impacts. This comprehensive approach addresses the current research gaps and provides valuable insights into the design and application of TE modules in various energy systems.

36 MATERIALS SCIENCE↗
DOE OSTI2025

HP-MDR: High-performance and Portable Data Refactoring and Progressive Retrieval with Advanced GPUs

Scientific applications produce vast amounts of data, posing grand challenges in the underlying data management and analytic tasks. Progressive compression is a promising way to address this problem, as it allows for on-demand data retrieval with significantly reduced data movement cost. However, most existing progressive methods are designed for CPUs, leaving a gap for them to unleash the power of today’s heterogeneous computing systems with GPUs.In this work, we propose HP-MDR, a high-performance and portable data refactoring and progressive retrieval framework for GPUs. Our contributions are four-fold: (1) We carefully optimize the bitplane encoding and lossless encoding, two key stages in progressive methods, to achieve high performance on GPUs; (2) We propose pipeline optimization and incorporate it with data refactoring and progressive retrieval workflows to further enhance the performance for large data process; (3) We leverage our framework to enable high-performance data retrieval with guaranteed error control for common Quantities of Interest; (4) We evaluate HP-MDR and compare it with state of the arts using five real-world datasets. Experimental results demonstrate that HP-MDR delivers an average 13.68 × and 6.31 × throughput in data refactoring and progressive retrieval tasks, respectively. It also leads to 11.22 × throughput for recomposing required data representations under Quantity-of-Interest error control and 6.04 × performance for the corresponding end-to-end data retrieval, when compared with state-of-the-art solutions.

Li, Yanliang [University of Oregon]↗
DOE OSTI2022

Composite fermion mass: Experimental measurements in ultrahigh quality two-dimensional electron systems

Composite fermions (CFs), exotic quasiparticles formed by pairing an electron and an even number of magnetic flux quanta, emerge at high magnetic fields in an interacting electron system, and can explain phenomena such as the fractional quantum Hall state (FQHS) and other many-body phases. CFs possess an effective mass (m CF ) whose magnitude is inversely related to the most fundamental property of a FQHS, namely its energy gap. Here we present here experimental measurements of m CF in ultrahigh quality two-dimensional electron systems confined to GaAs quantum wells of varying thickness. An advantage of measuring m CF over gap measurements is that mass values are insensitive to disorder and are therefore ideal for comparison with theoretical calculations, especially for high-order FQHS. Our data reveal that m CF increases with increasing well width, reflecting a decrease in the energy gap as the electron layer becomes thicker and the in-plane Coulomb energy softens. Comparing our measured masses with available theoretical results, we find significant quantitative discrepancies, highlighting that more rigorous and accurate calculations are needed to explain the experimental data.

36 MATERIALS SCIENCE↗
DOE OSTI2021

Unraveling the intrinsic atomic physics behind x-ray absorption line shifts in warm dense silicon plasmas

Here, we present a novel free-energy density functional theory (DFT)-based methodology for optical property calculations of warm dense matter to cover a wide range of thermodynamic conditions and photon energies including the entire x-ray range. It uses Mermin–Kohn–Sham density functional theory with exchange-correlation (XC) thermal effects taken into account via a fully temperature dependent generalized gradient approximation XC functional. The methodology incorporates a combination of the ab initio molecular dynamics (AIMD) snapshotted Kubo–Greenwood optic data with a single-atom in simulation cell calculations to close the photon energy gap between the L and K edges and extend the K-edge tail toward many-keV photon energies. This gap arises in the standard scheme due to a prohibitively large number of bands required for the Kubo–Greenwood calculations with AIMD snapshots. Kubo–Greenwood data on snapshots provide an accurate description of optic properties at low photon frequencies slightly beyond the L-edge and x-ray absorption near edges structure (XANES) spectra, while data from periodic calculations with single atoms cover the tail regions beyond the edges. To demonstrate its applicability to mid-Z materials where the standard DFT-based approach is not computationally feasible, we have applied it to opacity calculations of warm dense silicon plasmas. These first-principles calculations revealed a very interesting phenomon of red-to-blue shift in K–L (1s → 2p) and K-edge absorptions along both isotherm and isochore, which are absent in most continuum-lowering models of traditional plasma physics. This new physics phenomenon can be attributed to the underlying competition between the screening of deeply bound core electrons and the screening of outer-shell electrons caused by warm-dense-plasma conditions. We further demonstrate that the ratio of 1s → 2p to the K-edge x-ray absorptions can be used to characterize warm-dense-plasma conditions. Eventually, based on our absorption calculations, we have established a first-principles opacity table (FPOT) for silicon in a wide range of material densities and temperatures.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2021

Measurement of N* Cross-Sections from Single-Pion Electroproduction at low Q 2

Measurements of nucleon electroexcitation resonance cross-sections give unique insight into the dynamics of the strong interaction and our knowledge of Quantum Chromodynamics (QCD). The analysis of exclusive electroproduction data is an important tool in understanding the structure of the nucleon and its excited states. Exclusive channels give a clear way to extract the desired resonant contributions from non-resonant contributions. A gap in the helicity amplitudes extracted from single-pion cross-section data exists for virtual photon momentum transfers (Q 2 ) in the range of 1- 2 GeV 2 . Using data collected by the CEBAF Large Acceptance Spectrometer (CLAS) detector with a beam energy of 4.8 GeV, the reaction channel y*r → nπ + is used in this study to investigate the kinematic region covering W=[1.1,1.82)GeV and Q 2 = [1.1,3.5)GeV 2 . This work expands on the previous kinematic coverage for charged single-pion electroproduction and increases the angular coverage of previously extracted kinematic bins.

Tyler, Nicholas↗
DOE OSTI2024

Question-answering system extracts information on injection drug use from clinical notes

Background. Injection drug use (IDU) can increase mortality and morbidity. Therefore, identifying IDU early and initiating harm reduction interventions can benefit individuals at risk. However, extracting IDU behaviors from patients’ electronic health records (EHR) is difficult because there is no other structured data available, such as International Classification of Disease (ICD) codes, and IDU is most often documented in unstructured free-text clinical notes. Although natural language processing can efficiently extract this information from unstructured data, there are no validated tools. Methods. Here, to address this gap in clinical information, we design a question-answering (QA) framework to extract information on IDU from clinical notes for use in clinical operations. Our framework involves two main steps: (1) generating a gold-standard QA dataset and (2) developing and testing the QA model. We use 2323 clinical notes of 1145 patients curated from the US Department of Veterans Affairs (VA) Corporate Data Warehouse to construct the gold-standard dataset for developing and evaluating the QA model. We also demonstrate the QA model’s ability to extract IDU-related information from temporally out-of-distribution data. Results. Here, we show that for a strict match between gold-standard and predicted answers, the QA model achieves a 51.65% F1 score. For a relaxed match between the gold-standard and predicted answers, the QA model obtains a 78.03% F1 score, along with 85.38% Precision and 79.02% Recall scores. Moreover, the QA model demonstrates consistent performance when subjected to temporally out-of-distribution data. Conclusions. Our study introduces a QA framework designed to extract IDU information from clinical notes, aiming to enhance the accurate and efficient detection of people who inject drugs, extract relevant information, and ultimately facilitate informed patient care.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2023

A rule-free workflow for the automated generation of databases from scientific literature

Abstract In recent times, transformer networks have achieved state-of-the-art performance in a wide range of natural language processing tasks. Here we present a workflow based on the fine-tuning of BERT models for different downstream tasks, which results in the automated extraction of structured information from unstructured natural language in scientific literature. Contrary to existing methods for the automated extraction of structured compound-property relations from similar sources, our workflow does not rely on the definition of intricate grammar rules. Hence, it can be adapted to a new task without requiring extensive implementation efforts and knowledge. We test our data-extraction workflow by automatically generating a database for Curie temperatures and one for band gaps. These are then compared with manually curated datasets and with those obtained with a state-of-the-art rule-based method. Furthermore, in order to showcase the practical utility of the automatically extracted data in a material-design workflow, we employ them to construct machine-learning models to predict Curie temperatures and band gaps. In general, we find that, although more noisy, automatically extracted datasets can grow fast in volume and that such volume partially compensates for the inaccuracy in downstream tasks.

36 MATERIALS SCIENCE↗
DOE OSTI2026

Facilitating Data Collection of Maintenance Events to Populate the Hydrogen Component Reliability Database (HyCReD)

The Hydrogen Component Reliability Database (HyCReD) is a collaborative project between the National Renewable Energy Laboratory, the University of Maryland, and hydrogen stakeholders to improve safety and reliability for hydrogen facilities by implementing component reliability data taxonomies that support hydrogen infrastructure failure rate analysis. The project aims to quantify failure rates of hydrogen components through high-quality data collection and analysis on root causes and maintenance needed. HyCReD provides a common database for cataloging hydrogen component failures which exists for reliability research in many other mature industries [2]. The database fills a gap for the hydrogen community by providing a scientifically rigorous approach to quantitative risk assessment (QRA), prognostic health management (PHM), and reliability-centered maintenance (RCM) analysis. High level results will be aggregated and anonymized to protect company sensitive information; detailed results will be used to help address issues of hydrogen components. These advanced analytics will support accelerated deployment of hydrogen infrastructure by enabling better: design and safety of projects (safety codes and standards development), infrastructure reliability and cost (component failure rates, maintenance protocols), and component R&D needs (robust supply chain). A key to a successful HyCReD implementation is facilitating the ease of reporting and data quality in the database that can be used for analysis. Maintenance data was a previously identified gap in initial efforts to populate and validate the database taxonomies [3]. Collection of maintenance data will be instrumental in identifying failure modes and rates, identifying incipient component failures or reduced performance, cataloging best practices for maintenance routines and methods for prognostic health management, and quantifying the risk and effect of different failure modes. Several key priorities are identified for streamlined data collection to achieve quality and detailed failure data: Applicability, Ease of Use, Accessibility, and Information Security. The HyCReD team has now begun deployment of the database to several companies and groups that have signed non-disclosure agreements to facilitate the data collection of failures in industry hydrogen refueling station infrastructure. This paper will provide an update into the process of HyCReD deployment including the development of a coding guide for facility personnel to reference and ensure data quality and consistency from one station to another as well as implementation of contextually dependent data fields of system taxonomy and formatted entries to provide ease of use. The goal is to communicate the lessons learned from the roll-out to technicians and engineers in the field, and the addition of need for high level of security to protect all stakeholders.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗
DOE OSTI2023

Characterization and dynamics of the peculiar stream Jhelum: A tentative role for the Sagittarius dwarf galaxy

Stellar streams are a promising tool to study the Milky Way’s dark matter subhalo population, as interactions with subhalos are expected to leave visible imprints in the streams in the form of substructure. However, there may be other causes of substructure. Here we studied the kinematics and the unusual morphology of the stellar stream Jhelum. Using a combination of ground-based photometry and Gaia EDR3 astrometry, we characterized the morphology of Jhelum. We combined this new data with radial velocities from the literature to perform orbit integrations of the stream in static Galactic potentials. We also carried out N-body simulations in the presence of the Sagittarius dwarf galaxy. The new data reveal a previously unreported tertiary component in the stream, as well as several gaps and a kink-like feature in its narrow component. We find that for a range of realistic Galactic potentials, no single orbit is able to reproduce Jhelum’s radial velocity data entirely. A generic property of the orbital solutions is that they share a similar orbital plane to Sagittarius and this leads to repeated encounters with the stream. Using N-body simulations that include a massive Sagittarius, we explored its effect on Jhelum, and we show that these encounters are able to qualitatively reproduce the narrow and broad components in Jhelum, as well as create a tertiary component in some cases. We also find evidence that such encounters can result in an apparent increase in the velocity dispersion of the stream by a factor up to four due to overlapping narrow and broad components. Our findings suggest that the Jhelum stream is even more complex than once thought; however, its morphology and kinematics can tentatively be explained via the interactions with Sagittarius. In this scenario, the formation of Jhelum’s narrow and broad components occurs naturally, yet some of the smaller gap-like features remain to be explained.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2021

Assessment of the driving factors of CO 2 mitigation costs of household biogas systems in China: A LMDI decomposition with cost analysis model

China has been placing a substantial focus on biogas for reducing energy consumption and carbon dioxide (CO 2 ) emissions. The operation mode of biogas systems may make the CO 2 reduction target over-optimistic. There is limited research to investigate the influential factors that may be causing the gap between the actual and theoretical CO 2 reduction costs of biogas systems in China. In this research, by using field survey data of 209 biogas users and 489 non-biogas users from 19 villages in 2015, the gap between actual and theoretical unit CO 2 reduction cost is quantified at approximately 156 USD/t CO 2 . By employing the Logarithmic Mean Divisia Index I (LMDI) model, it is found that both the cost effect (48%) and the reduction effect (52%) contribute to the unit CO 2 reduction cost gap. Furthermore, four influential factors–household labor, accessibility to the energy resource, acceptance of biogas technology, and subsidy–significantly narrow the gap between actual and theoretical CO 2 reduction costs, while the levelized subsidy contributes to widening the gap. On average, biogas systems should be operated for at least four years and the substitution rate should be more than 67% in order to keep the gap between actual and theoretical CO 2 reduction costs under 50%.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗
DOE OSTI2026

From Bricks to Clicks: Mapping the White Space in Building Innovation

It is a critical national imperative to transform the buildings sector, yet innovation is impeded by deployment failures that leave promising technologies stranded. Conventional market reports and techno-economic analysis provide an insufficient understanding of markets and resource allocation for emerging building technologies. They omit crucial commercialization factors such as ecosystem maturity and adoption friction, where the coordinated participation of a network of suppliers, contractors, financiers, regulators, and integrators is required to scale solutions. This study addresses these gaps by introducing an evaluation framework grounded in front-line data from six years of the DOE's IMPEL incubator, comprising experience from 300 building-sector innovators and the adjacent, complex ecosystem. Our methodology synthesizes top-down market analysis with bottom-up, practitioner-level data across five megatrends: (M1) Affordable materials and industrialized construction; (M2) Healthy and efficient mechanical systems; (M3) Intelligent building operations; (M4) Buildings as grid assets; and (M5) High-density power and cooling for data centers and therein identify twelve "white space" technology opportunities. Next, we develop a multi-criteria scoring rubric to rank these opportunities based on parameters, i.e., Affordability, Quality of Life, Reliability, and Security, yielding composite ‘Demand’ and ‘Maturity’ indices. Our results indicate that the most significant white spaces may not be incremental products but a new class of ‘Ecosystem Enablers’, such as logistics platforms, orchestration layers, and automated compliance software that solve structural deployment gaps. This paper summarizes this transparent, evidence-based, practitioner-informed evaluation framework for policymakers and investors to re-evaluate policy and resource allocation and unlock scalable market transformation.

Singh, Reshma↗
DOE OSTI2021

Band structure dependent electronic localization in macroscopic films of single-chirality single-wall carbon nanotubes

Significant understanding has been achieved over the last few decades regarding chirality-dependent properties of single-wall carbon nanotubes (SWCNTs), primarily through single-tube studies. However, macroscopic manifestations of chirality dependence have been limited, especially in electronic transport, despite the fact that such distinct behaviors are needed for many applications of SWCNT-based devices. In addition, developing reliable transport theory is challenging since a description of localization phenomena in an assembly of nanoobjects requires precise knowledge of disorder on multiple spatial scales, particularly if the ensemble is heterogeneous. Here, we report an observation of pronounced chirality-dependent electronic localization in temperature and magnetic field dependent conductivity measurements on macroscopic films of single-chirality SWCNTs. The samples included large-gap semiconducting (6,5) and (10,3) films, narrow-gap semiconducting (7,4) and (8,5) films, and armchair metallic (6,6) films. Experimental data and theoretical calculations revealed Mott variable-range-hopping dominated transport in all samples, while localization lengths fall into three distinct categories depending on their band gaps. Armchair films have the largest localization length. Here our detailed analyses on electronic transport properties of single-chirality SWCNT films provide significant new insight into electronic transport in ensembles of nanoobjects, offering foundations for designing and deploying macroscopic SWCNT solid-state devices.

36 MATERIALS SCIENCE↗
DOE OSTI2021

Physics-constrained, low-dimensional models for magnetohydrodynamics: First-principles and data-driven approaches

Plasmas are highly nonlinear and multiscale, motivating a hierarchy of models to understand and describe their behavior. However, there is a scarcity of plasma models of lower fidelity than magnetohydrodynamics (MHD), although these reduced models hold promise for understanding key physical mechanisms, efficient computation, and real-time optimization and control. Galerkin models, obtained by projection of the MHD equations onto a truncated modal basis, and data-driven models, obtained by modern machine learning and system identification, can furnish this gap in the lower levels of the model hierarchy. This work develops a reduced-order modeling framework for compressible plasmas, leveraging decades of progress in projection-based and data-driven modeling of fluids. We begin by formalizing projection-based model reduction for nonlinear MHD systems. To avoid separate modal decompositions for the magnetic, velocity, and pressure fields, we introduce an energy inner product to synthesize all of the fields into a dimensionally consistent, reduced-order basis. Next, we obtain an analytic model by Galerkin projection of the Hall-MHD equations onto these modes. We illustrate how global conservation laws constrain the model parameters, revealing symmetries that can be enforced in data-driven models, directly connecting these models to the underlying physics. We demonstrate the effectiveness of this approach on data from high-fidelity numerical simulations of a three-dimensional spheromak experiment. Finally, this manuscript builds a bridge to the extensive Galerkin literature in fluid mechanics and facilitates future principled development of projection-based and data-driven models for plasmas.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2026

A ModEx Framework for Watershed Subsurface Investigation With Limited Geophysical Data Using Machine Learning and Hydrologic Modeling

Abstract Subsurface heterogeneity influences watershed hydrology strongly but remains difficult to characterize at catchment scales with sparse and costly field data. Geophysical surveys such as electromagnetic induction (EMI) provide local spatial subsurface images yet scaling them to watershed scales and converting EMI‐derived resistivity into hydraulic properties remains a challenge. We present a Model–Experiment (ModEx) framework that integrates limited EMI data with machine learning (ML) and hydrologic modeling to improve process representation and guide field investigations. Sparse EMI surveys were scaled to the catchment scale using a Random Forest model, and the resulting resistivity fields were combined with nearby borehole constraints to parameterize a hydrologic model. The EMI‐informed hydrological simulations improved predictions of streamflow sustained by subsurface flow and shallow saturation patterns. By combining EMI data and ML with hydrologic modeling, the ModEx framework guides future subsurface surveys, providing a transferable and efficient strategy for data–model integration across diverse watersheds. Plain Language Summary Mapping the underground network of soil and rock that controls water is essential for predicting floods and droughts, but seeing underground is difficult and expensive. We cannot drill everywhere, so scientists use geophysical tools to scan broad areas. There are two key challenges: these geophysical scans are often sparse across the whole watershed, and the geophysical data is hard to translate into water‐related properties. We used artificial intelligence to solve these problems. We taught a computer to find patterns linking the limited geophysical data to the land surface properties. This allowed it to fill in the gaps and create a complete, useful subsurface map for the entire watershed. This new map improves hydrologic simulations, leading to more accurate predictions of water movement in the watershed. It also helps scientists build better models with less data and generates a priority map showing where to measure next, making future investigations more efficient. Key Points Limited EMI scaled with ML improves catchment‐scale subsurface parameterization for hydrologic models The framework integrates hydrologic modeling with limited geophysical data to support subsurface investigation design ModEx framework offers a transferable data–model integration strategy that quantifies and reduces uncertainty guiding watershed studies

Chen, Hang↗
DOE OSTI2023

Observation of momentum-dependent charge density wave gap in a layered antiferromagnet ${\textrm{Gd}}{\textrm{Te}}_{3}$

Charge density wave (CDW) ordering has been an important topic of study for a long time owing to its connection with other exotic phases such as superconductivity and magnetism. The RTe 3 (R = rare-earth elements) family of materials provides a fertile ground to study the dynamics of CDW in van der Waals layered materials, and the presence of magnetism in these materials allows to explore the interplay among CDW and long range magnetic ordering. Here, we have carried out a high-resolution angle-resolved photoemission spectroscopy (ARPES) study of a CDW material GdTe 3 , which is antiferromagnetic below ~ 12 K, along with thermodynamic, electrical transport, magnetic, and Raman measurements. Our ARPES data show a two-fold symmetric Fermi surface with both gapped and ungapped regions indicative of the partial nesting. The gap is momentum dependent, maximum along $\overline{Γ}$ - $\overline{Z}$ and gradually decreases going towards $\overline{Γ}$ - $\overline{Z}$. Our study provides a platform to study the dynamics of CDW and its interaction with other physical orders in two- and three dimensions.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2025

Development of Predictive Model for Accurate Rupture Time from Multi-Axial Creep in Alloy 709 with Physics-Based Simulations

A physics-based model is developed to predict multiaxial creep behavior in Alloy 709 (A709), an advanced austenitic stainless steel intended for high-temperature applications such as Sodium Fast Reactors (SFRs). Compared to conventional stainless steels like 316H, A709 offers superior high-temperature performance; however, comprehensive data on its multiaxial creep response remain limited. To address this gap, a crystal plasticity finite element (CPFE) framework is used to simulate the deformation and failure mechanisms of A709 under multiaxial loading conditions. The model incorporates an extended Hu-Cocks dislocation creep formulation that accounts for precipitation effects, along with the Sham–Needleman model to capture grain boundary cavitation-driven failure. These advanced constitutive models enable a detailed understanding of the interplay between microstructural evolution and macroscopic creep response. Furthermore, the study evaluates the predictive accuracy of various effective stress measures in estimating creep rupture life, leveraging simulated multiaxial creep data. The findings provide critical insights into the applicability of different stress measures for engineering design and life prediction of A709 components operating under complex loading conditions. This work contributes to improving the reliability of high-temperature structural components by advancing predictive modeling capabilities for advanced austenitic steels.

Alloy 709↗
DOE OSTI2024

Efficient distributed continual learning for steering experiments in real-time

Deep learning has emerged as a powerful method for extracting valuable information from large volumes of data. However, when new training data arrives continuously (i.e., is not fully available from the beginning), incremental training suffers from catastrophic forgetting (i.e., new patterns are reinforced at the expense of previously acquired knowledge). Training from scratch each time new training data becomes available would result in extremely long training times and massive data accumulation. Rehearsal-based continual learning has shown promise for addressing the catastrophic forgetting challenge, but research to date has not addressed performance and scalability. To fill this gap, we propose an approach based on a distributed rehearsal buffer that efficiently complements data-parallel training on multiple GPUs to achieve high accuracy, short runtime, and scalability. It leverages a set of buffers (local to each GPU) and uses several asynchronous techniques for updating these local buffers in an embarrassingly parallel fashion, all while handling the communication overheads necessary to augment input minibatches using unbiased, global sampling. We further propose a generalization of rehearsal buffers to support both classification and generative learning tasks, as well as more advanced rehearsal strategies (notably Dark Experience Replay, leveraging knowledge distillation). We illustrate this approach with a real-life HPC streaming application from the domain of ptychographic image reconstruction. Furthermore, we run extensive experiments on up to 128 GPUs of the ThetaGPU supercomputer to compare our approach with baselines representative of training-from-scratch (the upper bound in terms of accuracy) and incremental training (the lower bound). Results show that rehearsal-based continual learning achieves a top-5 validation accuracy close to the upper bound, while simultaneously exhibiting a runtime close to the lower bound.

Asynchronous data management↗