Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “data needs”

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 1,423 records · Page 79
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

Designing alloys with process-mapping AI pre-trained on empirical knowledge

<span style="font-family: Calibri, sans-serif; font-size: 12pt;">Accelerated materials design should match the recent trends in the product development cycles. Materials data analytics can be used to significantly shorten development time of specialized alloys needed for next generation energy applications. However, it faces a challenge of scarce data available for training ML models. Incorporation of the domain knowledge into deep-learning graph structure via fuzzy pre-training and causal process imitation presents a viable approach to developing accurate data-driven models and reliable alloy design tools, with limited datasets. Artificial Intelligence (AI) was used in this study to incorporate such knowledge in the domain-specific computational tool, pyroMind. The tool provides not only novel design ideas but also their interpretation via physics and engineering concepts.</span>

Romanov, Vyacheslav↗
DOE OSTI2020

Generating synthetic cosmological data with GalSampler

As part of the effort to meet the needs of the Large Synoptic Survey Telescope Dark Energy Science Collaboration (LSST DESC) for accurate, realistically complex mock galaxy catalogues, in this work we have developed galsampler, an open-source python package that assists in generating large volumes of synthetic cosmological data. The key idea behind galsampler is to recast hydrodynamical simulations and semi-analytic models as physically motivated galaxy libraries. galsampler populates a new, larger volume halo catalogue with galaxies drawn from the baseline library; by using weighted sampling guided by empirical modelling techniques, galsampler inherits statistical accuracy from the empirical model and physically motivated complexity from the baseline library. We have recently used galsampler to produce the cosmoDC 2 extragalactic catalogue made for the LSST DESC Data Challenge 2 . Using cosmoDC 2 as a guiding example, we outline how galsampler can continue to support ongoing and near-future galaxy surveys such as the Dark Energy Survey, the Dark Energy Spectroscopic Instrument, WFIRST, and Euclid.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2023

Rapidly Changing Range Limits in a Warming World: Critical Data Limitations and Knowledge Gaps for Advancing Understanding of Mangrove Range Dynamics in the Southeastern USA

Climate change is altering species’ range limits and transforming ecosystems. For example, warming temperatures are leading to the range expansion of tropical, cold-sensitive species at the expense of their cold-tolerant counterparts. In some temperate and subtropical coastal wetlands, warming winters are enabling mangrove forest encroachment into salt marsh, which is a major regime shift that has significant ecological and societal ramifications. Here, we synthesized existing data and expert knowledge to assess the distribution of mangroves near rapidly changing range limits in the southeastern USA. We used expert elicitation to identify data limitations and highlight knowledge gaps for advancing understanding of past, current, and future range dynamics. Mangroves near poleward range limits are often shorter, wider, and more shrublike compared to their tropical counterparts that grow as tall forests in freeze-free, resource-rich environments. The northern range limits of mangroves in the southeastern USA are particularly dynamic and climate sensitive due to abundance of suitable coastal wetland habitat and the exposure of mangroves to winter temperature extremes that are much colder than comparable range limits on other continents. Thus, there is need for methodological refinements and improved spatiotemporal data regarding changes in mangrove structure and abundance near northern range limits in the southeastern USA. Advancing understanding of rapidly changing range limits is critical for foundation plant species such as mangroves, as it provides a basis for anticipating and preparing for the cascading effects of climate-induced species redistribution on ecosystems and the human communities that depend on their ecosystem services.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

Speed of Sound Measurements of Binary Mixtures of Difluoromethane (R-32) with 2,3,3,3-Tetrafluoropropene (R-1234yf) or trans-1,3,3,3-Tetrafluoropropene (R-1234ze(E)) Refrigerants

In this article, sound speed data measured using a dual-path pulse-echo instrument are reported for binary mixtures of difluoromethane (R-32) with 2,3,3,3-tetrafluoropropene (R-1234yf) or trans-1,3,3,3-tetrafluoropropene (R-1234ze(E)). The sound speed is reported at two compositions for each binary mixture of approximately (0.33/67) and (0.67/0.33) mole fraction at temperatures between 230 K and 345 K. Data are reported from pressures slightly above the bubble point to 12 MPa for R-32/1234yf mixtures to avoid potential polymerization reactions and to 53 MPa for the R-32/1234ze(E) mixtures. The mean uncertainty of the sound speed data are less than 0.1% of the measured value where uncertainties at individual state points range from 0.04% to 0.5% of the measured value as the conditions approach the mixture critical region. The reported data are compared to available Helmholtz-energy-explicit EOS included in REFPROP and all systems studied have average absolute deviations greater than 2%. The comparisons show that further adjustments to the mixture models are needed to provide a reasonable representation of the data within its experimental uncertainty.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

MODELING AND PARAMETRIC STUDY OF END-GAS AUTOIGNITION TO ALLOW THE REALIZATION OF ULTRA-LOW EMISSIONS, HIGH-EFFICIENCY HEAVY-DUTY SPARK-IGNITED NATURAL GAS ENGINES

Engine knock and misfire are barriers to pathways leading to high-efficiency Spark-Ignited (SI) Natural Gas (NG) engines. The general tendency to knock is highly dependent on engine operating conditions and the fuel reactivity. The problem is further complicated by the low emission limits and the wide range of chemical reactivity in pipeline-quality natural gas. Depending on the region and the source of the natural gas, its reactivity, described by its Methane Number (MN), which is analogous to the Octane Number for liquid SI fuels, can span from 65 to 95. In order to realize diesel-like efficiencies, SI NG engines must be designed to operate at high Brake Mean Effective Pressures (BMEP), near or beyond knock limits, over a wide range of fuel reactivity. This requires a deep understanding of the combustion-engine interactions pertaining to flame propagation and End-Gas Autoignition (EGAI), i.e., the autoignition of the unburned gas (end gas) ahead of the flame front. However, EGAI, if controlled, provides an opportunity to increase SI NG engine efficiency by increasing the combustion rate and the total fraction of burned fuel, mitigating the effects of the slow flame speeds characteristic of natural gas fuels, which generally reduce BMEP and increase unburned hydrocarbon emissions. For this reason, to realize diesel-like efficiencies and ultra-low emissions on SI NG engines, this work proposes the study of the main parameters influencing the modeling and prediction of NG EGAI to allow for its control. In this work, a novel EGAI detection and onset determination method was developed to reliably quantify EGAI for data analysis and engine control. The new method allowed the prediction of EGAI on SI NG engines without the need to use engine- and operating-condition-dependent thresholds and reduced the error in quantifying the fraction of the total energy released by the EGAI event by up to 40%pts. One- and three-dimensional engine models were then developed to study the engine/fuel interactions that lead to NG EGAI and its performance benefits. These models, although having decent agreement with experimental data, showed the need to account for NOx chemistry when predicting NG EGAI due to a consistently later prediction of the EGAI onset (~1.65 crank-angle degrees) and thus, a new reduced chemical mechanism for real NG fuels was developed containing NOx chemistry. The new reduced mechanism improved the EGAI onset prediction agreement to within ±0.5 crank-angle degrees and decreased simulation time during combustion by nearly 50% when using the further reduced AREIS50NOx chemical mechanism. These models were then used to study the role of NG composition on EGAI, evaluate the engine/fuel interactions leading to NG EGAI, and perform engine optimization while leveraging EGAI to increase thermal efficiency. Piston design optimization combined with a Controlled EGAI (C-EGAI) combustion mode allowed a Heavy-Duty (HD) SI NG engine to operate at diesel-like efficiencies, i.e., Brake Thermal Efficiency (BTE) ≥44%. Experimental and modeling data analysis revealed that earlier and faster heat release increases combustion efficiency by an average of 1%pts, increases work transferred to the piston resulting in a decrease in exhaust losses by 50% depending on the engine operating condition while slightly increasing heat losses. Finally, the simulation results revealed an opportunity to further enhance the BTE (up to 50%) by enabling C-EGAI combustion at leaner conditions, λ=1.4-1.6.

Bestel, Diego Bernardi↗
DOE OSTI2022

funcX: Federated Function as a Service for Science

Here, funcX is a distributed function as a service (FaaS) platform that enables flexible, scalable, and high performance remote function execution. Unlike centralized FaaS systems, funcX decouples the cloud-hosted management functionality from the edge-hosted execution functionality. funcX's endpoint software can be deployed, by users or administrators, on arbitrary laptops, clouds, clusters, and supercomputers, in effect turning them into function serving systems. funcX's cloud-hosted service provides a single location for registering, sharing, and managing both functions and endpoints. It allows for transparent, secure, and reliable function execution across the federated ecosystem of endpoints-enabling users to route functions to endpoints based on specific needs. funcX uses containers (e.g., Docker, Singularity, and Shifter) to provide common execution environments across endpoints. funcX implements various container management strategies to execute functions with high performance and efficiency on diverse funcX endpoints. funcX also integrates with an in-memory data store and Globus for managing data that may span endpoints. We motivate the need for funcX, present our prototype design and implementation, and demonstrate, via experiments on two supercomputers, that funcX can scale to more than 130000 concurrent workers. We show that funcX's container warming-aware routing algorithm can reduce the completion time for 3,000 functions by up to 61% compared to a randomized algorithm and the in-memory data store can speed up data transfers by up to 3x compared to a shared file system.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

Data Architecture and Analytics Requirements for Artificial Intelligence and Machine Learning Applications to Achieve Condition-Based Maintenance

This report identified some of the important requirements that needs to be taken into consideration as part of the data evolution for the CBM application of a CWS in a NPP. In the data evolution process, the information is converted into insight leading into actions using advancements in AI/ML technologies. A notion of RESET AI: design, development, deployment, and operation principals are introduced to lifecycle of AI technologies. Towards the end of the report, we discussed how this CBM can be realized in a SDE. As path forward, this report lays the foundation for developing a more detailed industry guidance supporting data evolution for other plant applications like operations and plant support. These would be developed as part of ongoing research in the fiscal year 2023.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2021

Fission Product Yield Data Adjustment in a Prototype Version of TSURFER

The TSURFER (Tool for Sensitivity/Uncertainty analysis of Response Functionals using Experimental Results) module of Oak Ridge National Laboratory’s (ORNL’s) SCALE code system has been updated to perform nuclear data adjustments for fixed-source irradiation/depletion problems. TSURFER uses a generalized linear least squares (GLLS) approach to consolidate a prior set of measured responses and corresponding calculated values to create the most self-consistent set of nuclear data. Traditionally, TSURFER adjustments have been performed for multigroup nuclear data such as reaction cross sections. In this work, TSURFER is expanded to perform adjustments to independent fission product yields and branching ratios that need equality constraints. To preserve equality constraints after the data adjustment procedure, an updated GLLS formulation includes a new Lagrange multiplier that forces data adjustment to sum to 0 for a given fission yield/branching ratio parent. A test problem illustrates that the newly updated TSURFER module satisfies the required constraint that adjustments for fission yield data sum to 0.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2023

Fusing time-varying mosquito data and continuous mosquito population dynamics models

Climate change is arguably one of the most pressing issues affecting the world today and requires the fusion of disparate data streams to accurately model its impacts. Mosquito populations respond to temperature and precipitation in a nonlinear way, making predicting climate impacts on mosquito-borne diseases an ongoing challenge. Data-driven approaches for accurately modeling mosquito populations are needed for predicting mosquito-borne disease risk under climate change scenarios. Many current models for disease transmission are continuous and autonomous, while mosquito data is discrete and varies both within and between seasons. This study uses an optimization framework to fit a non-autonomous logistic model with periodic net growth rate and carrying capacity parameters for 15 years of daily mosquito time-series data from the Greater Toronto Area of Canada. The resulting parameters accurately capture the inter-annual and intra-seasonal variability of mosquito populations within a single geographic region, and a variance-based sensitivity analysis highlights the influence each parameter has on the peak magnitude and timing of the mosquito season. This method can easily extend to other geographic regions and be integrated into a larger disease transmission model. This method addresses the ongoing challenges of data and model fusion by serving as a link between discrete time-series data and continuous differential equations for mosquito-borne epidemiology models.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Results of the Photometric LSST Astronomical Time-series Classification Challenge (PLAsTiCC)

Abstract Next-generation surveys like the Legacy Survey of Space and Time (LSST) on the Vera C. Rubin Observatory (Rubin) will generate orders of magnitude more discoveries of transients and variable stars than previous surveys. To prepare for this data deluge, we developed the Photometric LSST Astronomical Time-series Classification Challenge (PLAsTiCC), a competition that aimed to catalyze the development of robust classifiers under LSST-like conditions of a nonrepresentative training set for a large photometric test set of imbalanced classes. Over 1000 teams participated in PLAsTiCC, which was hosted in the Kaggle data science competition platform between 2018 September 28 and 2018 December 17, ultimately identifying three winners in 2019 February. Participants produced classifiers employing a diverse set of machine-learning techniques including hybrid combinations and ensemble averages of a range of approaches, among them boosted decision trees, neural networks, and multilayer perceptrons. The strong performance of the top three classifiers on Type Ia supernovae and kilonovae represent a major improvement over the current state of the art within astronomy. This paper summarizes the most promising methods and evaluates their results in detail, highlighting future directions both for classifier development and simulation needs for a next-generation PLAsTiCC data set.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2020

Technical Gaps in Hanford High-Level Waste Solids Settling Behavior and Settling Time Evaluation for Direct Feed High-Level Waste (DFHLW) Operations

Settling of high-level waste (HLW) solids in process vessels is a key conceptual process step in providing HLW feed directly to the Hanford Waste Treatment and Immobilization Plant (WTP) HLW Vitrification Facility. Direct Feed High-Level Waste (DFHLW) is a potential approach to initiating HLW vitrification prior to completing of the WTP Pretreatment Facility. Settling would be used with subsequent supernatant decant to concentrate HLW feed. To support planning for DFHLW, Washington River Protection Solutions (WRPS) requested support from the Pacific Northwest National Laboratory to evaluate the current data set available to predict the time needed for HLW solids to settle, to identify gaps in the understanding and predictive capability of HLW solids waste settling times, and to provide scoping estimates of the potential settling time. Eight technical gaps were identified for predicting settling times and characteristics of the formed sediment layers including: Gap 1: In-Tank Settling Rates Faster than Settling of Laboratory Samples, Gap 2: Effect of Sludge Leaching/Washing on Predicted Settling Times, Gap 3: Predicting Waste Settling from Waste Chemistry (Waste Type), Gap 4: Predicting Waste Settling from Particle Size and Density Distributions (PSDDs), Gap 5: Insufficient Laboratory and In-Tank Settling Data to Represent Hanford Waste, Gap 6: Methods for Real-Time, In-Tank Tracking of Settling, Gap 7: Prediction of Sediment Erosion Resistance as a Function of Settling Time, and Gap 8: Prediction of Sediment Solids Content as a Function of Settling Time. In addition to the data gaps, an overarching observation of the settling rate and settled layer data is the significant variation in behavior. At similar solids concentrations, settling rates can vary by as much as 3 orders of magnitude depending on the source waste tank, and significantly different settling rates are noted between laboratory and in situ tests for the same waste tank. The range of average solids concentration in existing HLW sediment, which may have been quiescent for decades, can vary from less than 7 wt% to greater than 74 wt% solids. The shear strengths (or yield stresses) measured on laboratory samples range from less than 27 Pa to greater than 6400 Pa. These variations can challenge process planning for the application of a settle/decant process for DFHLW. This report describes the significance of the gaps to the settle/decant process and presents uncertainties by way of examples. Potential technical approaches for resolving these gaps are described and the estimated difficulty in resolving these gaps is evaluated. Based on the significance of the gap and the difficulty of resolution, recommendations are made to address specific gaps. Scoping estimates of the potential settling times for DFHLW solids have been made based on the existing data set with its associated gaps. Depending on the process vessel depth and final sediment concentration, substantial fractions of the scoping estimate results for settling times for characterized HLW exceed the 2-week period that has been previously assumed for process planning. There is also significant disparity, potentially greater than a factor of 5000 difference, in the estimated settling times depending on process vessel depth and final sediment solid concentration. This variation in results underscores the significance of the identified gaps and uncertainties with respect to process planning for utilizing settle/decant operations for DFHLW.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗
DOE OSTI2025

Neutrino-nucleus cross section impacts on neutrino oscillation measurements

The challenges in neutrino-nucleus cross section modeling and its impact on neutrino oscillation experiments are widely recognized. However, a comprehensive and theoretically robust estimation of cross section uncertainties has been lacking, and few studies have quantitatively examined their impact on oscillation measurements. In this work, we evaluate the effect of cross section uncertainties on oscillation parameters using setups inspired by NOvA and DUNE. To characterize these uncertainties, we adopt multiple neutrino-nucleus event generators and simulate a realistic experimental procedure that incorporates near-detector data and near-to-far-detector extrapolation. Our results confirm that cross section uncertainties cannot significantly bias oscillation results in current statistics-dominated experiments like NOvA. However, they could lead to substantial bias for future systematics-dominated experiments like DUNE, even when near-detector data are employed to mitigate uncertainties. These findings underscore the need for further studies on the quantitative impacts of cross section modeling, improved strategies to utilize near-detector data and the PRISM concept, and more robust cross section models to optimize the success of future experiments.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2024

Phase Diagrams of Alloys and Their Hydrides via On-Lattice Graph Neural Networks and Limited Training Data

Efficient prediction of sampling-intensive thermodynamic properties is needed to evaluate material performance and permit high-throughput materials modeling for a diverse array of technology applications. To alleviate the prohibitive computational expense of high-throughput configurational sampling with density functional theory (DFT), surrogate modeling strategies like cluster expansion are many orders of magnitude more efficient but can be difficult to construct in systems with high compositional complexity. We therefore employ minimal-complexity graph neural network models that accurately predict and can even extrapolate to out-of-train distribution formation energies of DFT-relaxed structures from an ideal (unrelaxed) crystallographic representation. This enables the large-scale sampling necessary for various thermodynamic property predictions that may otherwise be intractable and can be achieved with small training data sets. Two exemplars, optimizing the thermodynamic stability of low-density high-entropy alloys and modulating the plateau pressure of hydrogen in metal alloys, demonstrate the power of this approach, which can be extended to a variety of materials discovery and modeling problems.

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

Uncertainty Quantification of Global Net Methane Emissions From Terrestrial Ecosystems Using a Mechanistically Based Biogeochemistry Model

Quantification of methane (CH4) emissions from wetlands and its sinks from uplands is still fraught with large uncertainties. Here, a methane biogeochemistry model was revised, parameterized, and verified for various wetland ecosystems across the globe. The model was then extrapolated to the global scale to quantify the uncertainty induced from four different types of uncertainty sources including parameterization, wetland type distribution, wetland area distribution, and meteorological input. We found that global wetland emissions are 212 ± 62 and 212 ± 32 Tg CH4 year -1 (1Tg = 10 12 g) due to uncertain parameters and wetland type distribution, respectively, during 2000–2012. Using two wetland distribution data sets and three sets of climate data, the model simulations indicated that the global wetland emissions range from 186 to 212 CH 4 year -1 for the same period. The parameters were the most significant uncertainty source. After combining the global methane consumption in the range of -34 to -46 Tg CH 4 year -1 , we estimated that the global net land methane emissions are 149–176 Tg CH 4 year -1 due to uncertain wetland distribution and meteorological input. Spatially, the northeast United States and Amazon were two hotspots of methane emission, while consumption hotspots were in the Eastern United States and eastern China. During 1950–2016, both wetland emissions and upland consumption increased during El Niño events and decreased during La Niña events. This study highlights the need for more in situ methane flux data, more accurate wetland type, and area distribution information to better constrain the model uncertainty.

58 GEOSCIENCES↗
DOE OSTI2021

Intermediate scattering functions of a rigid body monoclonal antibody protein in solution studied by dissipative particle dynamic simulation

In the past decade, there was increased research interest in studying internal motions of flexible proteins in solution using Neutron Spin Echo (NSE) as NSE can simultaneously probe the dynamics at the length and time scales comparable to protein domain motions. However, the collective intermediate scattering function (ISF) measured by NSE has the contributions from translational, rotational, and internal motions, which are rather complicated to be separated. Widely used NSE theories to interpret experimental data usually assume that the translational and rotational motions of a rigid particle are decoupled and independent to each other. To evaluate the accuracy of this approximation for monoclonal antibody (mAb) proteins in solution, dissipative particle dynamic computer simulation is used here to simulate a rigid-body mAb for up to about 200 ns. The total ISF together with the ISFs due to only the translational and rotational motions as well as their corresponding effective diffusion coefficients is calculated. The aforementioned approximation introduces appreciable errors to the calculated effective diffusion coefficients and the ISFs. For the effective diffusion coefficient, the error introduced by this approximation can be as large as about 10% even though the overall agreement is considered reasonable. Thus, we need to be cautious when interpreting the data with a small signal change. In addition, the accuracy of the calculated ISFs due to the finite computer simulation time is also discussed.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2024

Comparing gas composition from fast pyrolysis of live foliage measured in bench-scale and fire-scale experiments

Background: Fire models have used pyrolysis data from oxidising and non-oxidising environments for flaming combustion. In wildland fires pyrolysis, flaming and smouldering combustion typically occur in an oxidising environment (the atmosphere). Aims: Using compositional data analysis methods, determine if the composition of pyrolysis gases measured in non-oxidising and ambient (oxidising) atmospheric conditions were similar. Methods: Permanent gases and tars were measured in a fuel-rich (non-oxidising) environment in a flat flame burner (FFB). Permanent and light hydrocarbon gases were measured for the same fuels heated by a fire flame in ambient atmospheric conditions (oxidising environment). Log-ratio balances of the measured gases common to both environments (CO, CO 2 , CH 4 , H 2 , C 6 H 6 O (phenol), and other gases) were examined by principal components analysis (PCA), canonical discriminant analysis (CDA) and permutational multivariate analysis of variance (PERMANOVA). Key results: Mean composition changed between the non-oxidising and ambient atmosphere samples. PCA showed that flat flame burner (FFB) samples were tightly clustered and distinct from the ambient atmosphere samples. CDA found that the difference between environments was defined by the CO-CO 2 log-ratio balance. PERMANOVA and pairwise comparisons found FFB samples differed from the ambient atmosphere samples which did not differ from each other. Conclusion: Relative composition of these pyrolysis gases differed between the oxidising and non-oxidising environments. This comparison was one of the first comparisons made between bench-scale and field scale pyrolysis measurements using compositional data analysis. Implications: These results indicate the need for more fundamental research on the early time-dependent pyrolysis of vegetation in the presence of oxygen.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

Genesis Mission Data cards

As data-intensive research and artificial intelligence become central to DOE mission science, the need for machine-actionable dataset documentation has grown accordingly. However, many DOE-aligned communities, including the Office of Science, NNSA, and cross-laboratory collaborations, have developed independent metadata practices. This fragmentation creates friction for discovery, federation, and reuse across programs. To address these challenges, this talk introduces the Genesis Data Card: a shared metadata artifact developed in collaboration with a broad DOE community (Jefferson Lab and the National Lab of the Rockies, Oak Ridge, Sandia, Idaho, Berkeley, and Los Alamos). The Genesis Data Card aims to standardize dataset documentation across DOE-aligned initiatives while remaining extensible to discipline-specific needs. This talk will describe the data card template and the supporting code to validate completed data cards, using a companion LinkML schema. I'll walk through the design decisions behind the template, its alignment with existing standards, its treatment of sensitivity and governance metadata, and the phased roadmap toward lifecycle-integrated "xCards" that support autonomous discovery and reuse. The talk closes with current gaps, ongoing work, and how others can contribute datasets and feedback to the shared repository.

McSpadden, Helen [Thomas Jefferson National Accele↗
DOE OSTI2020

How Low Can You Go? Using Synthetic 3D Imagery to Drastically Reduce Real-World Training Data for Object Detection

Deep convolutional neural networks (DCNNs) currently provide state-of-the-art performance on image classification and object detection tasks, and there are many global security mission areas where such models could be extremely useful. Crucially, the success of these models is driven in large part by the widespread availability of high-quality open source data sets such as Image Net, Common Objects in Context (COCO), and KITTI, which contain millions of images with thousands of unique labels. However, global security relevant objects-of-interest can be difficult to obtain: relevant events are low frequency and high consequence; the content of relevant images is sensitive; and adversaries and proliferators seek to obscure their activities. For these cases where exemplar data is hard to come-by, even fine-tuning an existing model with available data can be effectively impossible. Recent work demonstrated that models can be trained using a combination of real-world and synthetic images generated from 3D representations; that such models can exceed the performance of models trained using real-world data alone; and that the generated images need not be perfectly realistic (Tremblay, et al., 2018). However, this approach still required hundreds to thousands of real-world images for training and fine tuning, which for sparse, global security-relevant datasets can be an unrealistic hurdle. In this research, we validate the performance and behavior of DCNN models as we drive the number of real-world images used for training object detection tasks down to a minimal set. We perform multiple experiments to identify the best approach to train DCNNs from an extremely small set of real-world images. In doing so, we: Develop state-of-the-art, parameterized 3D models based on real-world images and sample from their parameters to increase the variance in synthetic image training data; Use machine learning explainability techniques to highlight and correct through targeted training the biases that result from training using completely synthetic images; and Validate our results by comparing the performance of the models trained on synthetic data to one another, and to a control model created by fine-tuning an existing ImageNet-trained model with a limited number (hundreds) of real-world images.

97 MATHEMATICS AND COMPUTING↗