SEARCH · Engineering Papers

Results for “Classification bias”

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.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

158 records · Page 9

All sources NASA NTRS DOE OSTI

DOE OSTI2022

TULIP: An RNA-seq-based Primary Tumor Type Prediction Tool Using Convolutional Neural Networks

Background: With cancer as one of the leading causes of death worldwide, accurate primary tumor type prediction is critical in identifying genetic factors that can inhibit or slow tumor progression. There have been efforts to categorize primary tumor types with gene expression data using machine learning, and more recently with deep learning, in the last several years. Methods In this paper, we developed four 1-dimensional (1D) Convolutional Neural Network (CNN) models to classify RNA-seq count data as one of 17 highly represented primary tumor types or 32 primary tumor types regardless of imbalanced representation. Additionally, we adapted the models to take as input either all Ensembl genes (60,483) or protein coding genes only (19,758). Unlike previous work, we avoided selection bias by not filtering genes based on expression values. RNA-seq count data expressed as FPKM-UQ of 9,025 and 10,940 samples from The Cancer Genome Atlas (TCGA) were downloaded from the Genomic Data Commons (GDC) corresponding to 17 and 32 primary tumor types respectively for training and validating the models. Results: All 4 1D-CNN models had an overall accuracy of 94.7% to 97.6% on the test dataset. Further evaluation indicates that the models with protein coding genes only as features performed with better accuracy compared to the models with all Ensembl genes for both 17 and 32 primary tumor types. For all models, the accuracy by primary tumor type was above 80% for most primary tumor types. Conclusions: We packaged all 4 models as a Python-based deep learning classification tool called TULIP (TUmor CLassIfication Predictor) for performing quality control on primary tumor samples and characterizing cancer samples of unknown tumor type. Further optimization of the models is needed to improve the accuracy of certain primary tumor types.

Jones, Sara↗

NASA NTRS2021

Three-dimensional estimation of deciduous forest canopy structure and leaf area using multi-directional, leaf-on and leaf-off airborne lidar data

Airborne laser scanning (ALS) has been widely used to map gap probability and leaf area index (LAI) distribution at plot and landscape scales. As an indirect measurement, most ALS methods to estimate LAI combine waveform or point density information with supporting field measurements such as the leaf angle distribution, gap probability, or direct LAI measures. The development of a more independent estimation approach would facilitate more widespread use of existing ALS data to investigate patterns of forest structure and build realistic 3-D vegetation scenes to simulate remote sensing imagery and energy balance. Here, we develop a data processing workflow (named PVlad) using ALS point cloud apparent reflectance to estimate LAI and voxel-based leaf area density (LAD), aiming to reduce the need for associated field measurements such as the gap probability. The adaptation of the path volume (PV) concept derived from apparent reflectance integrates information from multi-directional ALS pulses, and quantifies the percentage exploration of each voxel for classification and occlusion correction, such that rigorous volumetric sampling approaches can be developed to derive LAI and LAD. The PVlad workflow was applied to discrete-return lidar data (Riegl VQ480i) acquired by NASA Goddard's LiDAR, Hyperspectral and Thermal Imager (G-LiHT) Airborne Imager during leaf-on (summer) and leaf-off (spring) conditions at the Smithsonian Environmental Research Center (SERC). The estimates of LAI and LAD captured structural differences between mature, logged, and intermediate-aged stands over eight deciduous forest plots. The derived LAI values were compared to field litter collection measurements, and the derived LAD vertical distribution was compared to the output of the VoxLAD model using terrestrial laser scan (TLS) field survey data. Using voxel sizes ranging from 0.5 m to 5 m, overall LAI estimation showed linear fitting coefficient bias and for 1 and 2 m voxel sizes, and vertical LAD distribution showed strong correlation with and for 0.5 and 1m voxel sizes. For every forest stand, upper-canopy LAD had a low variance for voxel sizes of ≤ . Application of PVlad to the G-LiHT and other similar ALS data archives enables the development of fine-resolution LAI map products, including voxelization of LAD for ecosystem science and radiative transfer simulations of remote sensing imagery or surface energy balance.

Tiangang Yin↗

NASA NTRS1997

Cloud Liquid Water Path Comparisons from Passive Microwave and Solar Reflectance Satellite Measurements: Assessment of Sub-Field-of-View Cloud Effects in Microwave Retrievals

Satellite observations of the cloud liquid water path (LWP) are compared from special sensor microwave imager (SSM/I) measurements and GOES 8 imager solar reflectance (SR) measurements to ascertain the impact of sub-field-of-view (FOV) cloud effects on SSM/I 37 GHz retrievals. The SR retrievals also incorporate estimates of the cloud droplet effective radius derived from the GOES 8 3.9-micron channel. The comparisons consist of simultaneous collocated and full-resolution measurements and are limited to nonprecipitating marine stratocumulus in the eastern Pacific for two days in October 1995. The retrievals from these independent methods are consistent for overcast SSM/I FOVS, with RMS differences as low as 0.030 kg/sq m, although biases exist for clouds with more open spatial structure, where the RMS differences increase to 0.039 kg/sq m. For broken cloudiness within the SSM/I FOV the average beam-filling error (BFE) in the microwave retrievals is found to be about 22% (average cloud amount of 73%). This systematic error is comparable with the average random errors in the microwave retrievals. However, even larger BFEs can be expected for individual FOVs and for regions with less cloudiness. By scaling the microwave retrievals by the cloud amount within the FOV, the systematic BFE can be significantly reduced but with increased RMS differences of O.046-0.058 kg/sq m when compared to the SR retrievals. The beam-filling effects reported here are significant and are expected to impact directly upon studies that use instantaneous SSM/I measurements of cloud LWP, such as cloud classification studies and validation studies involving surface-based or in situ data.

Greenwald, Thomas J.↗

DOE OSTI2020

GCAM v4.3 SSP-RCP-GCM Output Products

Supplemental dataset to: Graham N.T., M.I. Hejazi, M. Chen, E. Davies, J.A. Edmonds, S.H. Kim, and S. Turner, et al. 2020. "Humans drive future water scarcity changes across all Shared Socioeconomic Pathways." Environmental Research Letters 15, no. 1:Article No. 014007. PNNL-SA-151297. doi:10.1088/1748-9326/ab639b This analysis makes deterministic classifications of how, when water resources are limited and constrained, the human and climate systems interact. The relative effects of both systems on water scarcity are quantified at global and basin scales across 15 global futures that include five different socioeconomic conditions (the Shared Socioeconomic Pathways, SSPs) and four different climatic conditions (the Representative Concentration Pathways, RCPs). These 15 scenarios are used to first analyze a 'Human Alone' component by isolating the human impact on future scarcity. This is accomplished by holding all climate variables to their 2005 levels while altering socioeconomic growth and technological change. General circulation model (GCM) derived climate impacts for five models are applied to establish 75 'Human and Climate' scenarios which allows for the quantification of climate impacts. Climatic impacts to water supply, agricultural productivity and change, hydropower availability, and building energy demands are applied from 5 different bias-corrected GCMs to make a suite of 75 climate runs. By subtracting the human derived impacts from the 'Human Alone' scenarios, from the 'Human and Climate' scenarios the climate impact is isolated. First, the drivers of future water scarcity are evaluated by isolating the impacts that both humans and climate have while accounting for feedbacks between humans, energy, and land. Secondly, the simultaneous impacts that human and climate systems have on water scarcity are analyzed by determining whether each system is increasing or decreasing scarcity in all global water basins. Below we describe the GCAM model, the scenario components, climate derived impacts, and the calculations of scarcity changes and attribution. This study uses the global change assessment model (GCAM), with inclusions of water constraints to both renewable and nonrenewable sources of water, to investigate the relative contributions of climate and human systems on water scarcity regionally and globally under a wide range of scenarios. GCAM links socioeconomics, the energy system, land-use change, climate, and the water sector. GCAM is a market-equilibrium model that allows for prices to be adjusted within each time step to ensure that the supply and demand of goods and services remains equilibrated at each time step allowing for simultaneous market clearing across sectors. This study accounts for a limited supply of water by employing cost resource curves across all 235 basins that follow a logit formulation to determine the share of each water source (renewable surface water, nonrenewable groundwater, and desalinated water) needed to meet the water demands within all basins. As depletion of various water sources increases the extraction price increases, which leads to compounding price increases on the goods and services that require higher-priced water sources.

99 GENERAL AND MISCELLANEOUS↗

DOE OSTI2024

Multiple-charging effects on the CCN activity and hygroscopicity of surrogate black carbon particles

Accurate measurements of cloud condensation nuclei (CCN) activity and hygroscopicity of black carbon (BC)-containing particles are particularly important because of the positive climate forcing from these particles. Such measurements are typically conducted on particles selected by a Differential Mobility Analyzer (DMA), which in addition to singly charged particles transmits multiply charged larger particles that have the same electrical mobility. These larger particles activate at lower supersaturations than the singly charged particles, biasing measurements and resulting in overestimation of CCN activity and hygroscopicity parameter (κ). Here, we measure the CCN activity and determine κ for different BC surrogates with electrical mobility diameters from 100 to 200 nm selected 1) only by electrical mobility with a DMA, and 2) by both electrical mobility and mass using a DMA and a Centrifugal Particle Mass Analyzer (CPMA), thus allowing selection of only singly charged particles. We demonstrate the use of the DMA-CPMA system in resolving biases caused by multiply charged particles, and we show that the effect of multiple charging on the CCN activity of the BC particles is strongly influenced by morphology dispersion, i.e., the variability due to the range of morphologies of particles that have the same electrical mobility and mass. Finally, our findings show that electrical mobility-based methods alone are unlikely to lead to accurate results in measurements of CCN activation and hygroscopicity of BC particles, even for those with a more compact morphology.

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2021

Unsupervised machine learning for unbiased chemical classification in X-ray absorption spectroscopy and X-ray emission spectroscopy

Here we report a comprehensive computational study of unsupervised machine learning for extraction of chemically relevant information in X-ray absorption near edge structure (XANES) and in valence-to-core X-ray emission spectra (VtC-XES) for classification of a broad ensemble of sulphorganic molecules. By progressively decreasing the constraining assumptions of the unsupervised machine learning algorithm, moving from principal component analysis (PCA) to a variational autoencoder (VAE) to t-distributed stochastic neighbour embedding (t-SNE), we find improved sensitivity to steadily more refined chemical information. Surprisingly, when embedding the ensemble of spectra in merely two dimensions, t-SNE distinguishes not just oxidation state and general sulphur bonding environment but also the aromaticity of the bonding radical group with 87% accuracy as well as identifying even finer details in electronic structure within aromatic or aliphatic sub-classes. We find that the chemical information in XANES and VtC-XES is very similar in character and content, although they unexpectedly have different sensitivity within a given molecular class. We also discuss likely benefits from further effort with unsupervised machine learning and from the interplay between supervised and unsupervised machine learning for X-ray spectroscopies. Our overall results, i.e., the ability to reliably classify without user bias and to discover unexpected chemical signatures for XANES and VtC-XES, likely generalize to other systems as well as to other one-dimensional chemical spectroscopies.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2024

Theory for Equivariant Quantum Neural Networks

Quantum neural network architectures that have little to no inductive biases are known to face trainability and generalization issues. Inspired by a similar problem, recent breakthroughs in machine learning address this challenge by creating models encoding the symmetries of the learning task. This is materialized through the usage of equivariant neural networks the action of which commutes with that of the symmetry. In this work, we import these ideas to the quantum realm by presenting a comprehensive theoretical framework to design equivariant quantum neural networks (EQNNs) for essentially any relevant symmetry group. We develop multiple methods to construct equivariant layers for EQNNs and analyze their advantages and drawbacks. Our methods can find unitary or general equivariant quantum channels efficiently even when the symmetry group is exponentially large or continuous. As a special implementation, we show how standard quantum convolutional neural networks (QCNNs) can be generalized to group-equivariant QCNNs where both the convolution and pooling layers are equivariant to the symmetry group. We then numerically demonstrate the effectiveness of a S U ( 2 ) -equivariant QCNN over symmetry-agnostic QCNN on a classification task of phases of matter in the bond-alternating Heisenberg model. Our framework can be readily applied to virtually all areas of quantum machine learning. Lastly, we discuss about how symmetry-informed models such as EQNNs provide hopes to alleviate central challenges such as barren plateaus, poor local minima, and sample complexity. Published by the American Physical Society 2024

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

NASA NTRS1996

Fossil preservation and the stratigraphic ranges of taxa

The incompleteness of the fossil record hinders the inference of evolutionary rates and patterns. Here, we derive relationships among true taxonomic durations, preservation probability, and observed taxonomic ranges. We use these relationships to estimate original distributions of taxonomic durations, preservation probability, and completeness (proportion of taxa preserved), given only the observed ranges. No data on occurrences within the ranges of taxa are required. When preservation is random and the original distribution of durations is exponential, the inference of durations, preservability, and completeness is exact. However, reasonable approximations are possible given non-exponential duration distributions and temporal and taxonomic variation in preservability. Thus, the approaches we describe have great potential in studies of taphonomy, evolutionary rates and patterns, and genealogy. Analyses of Upper Cambrian-Lower Ordovician trilobite species, Paleozoic crinoid genera, Jurassic bivalve species, and Cenozoic mammal species yield the following results: (1) The preservation probability inferred from stratigraphic ranges alone agrees with that inferred from the analysis of stratigraphic gaps when data on the latter are available. (2) Whereas median durations based on simple tabulations of observed ranges are biased by stratigraphic resolution, our estimates of median duration, extinction rate, and completeness are not biased.(3) The shorter geologic ranges of mammalian species relative to those of bivalves cannot be attributed to a difference in preservation potential. However, we cannot rule out the contribution of taxonomic practice to this difference. (4) In the groups studied, completeness (proportion of species [trilobites, bivalves, mammals] or genera [crinoids] preserved) ranges from 60% to 90%. The higher estimates of completeness at smaller geographic scales support previous suggestions that the incompleteness of the fossil record reflects loss of fossiliferous rock more than failure of species to enter the fossil record in the first place.

NASA Program Exobiology↗

NASA NTRS2004

Top-of-Atmosphere Shortwave Broadband Observed Radiance and Estimated Irradiance over Polar Regions from Clouds and the Earth's Radiant Energy System (CERES) Instruments on Terra

Empirical angular distribution models for estimating top-of-atmosphere shortwave irradiances from radiance measurements over permanent snow, fresh snow and sea ice are developed using CERES measurements on Terra. Permanent snow angular distribution models depend on cloud fraction, cloud optical thickness, and snow brightness. Fresh snow and sea ice angular distribution models depend on snow and sea ice fraction, cloud fraction, cloud optical thickness, and snow and ice brightness. These classifications lead to 10 scene types for permanent snow and 25 scene types for fresh snow and sea ice. The average radiance over clear-sky permanent snow is more isotropic with satellite viewing geometry than that over overcast permanent snow. On average, the albedo of clear-sky permanent snow varies from 0.65 to 0.68 for solar zenith angles between 60$logical and\circ$ and 80 deg, while the corresponding albedo of overcast scenes varies from 0.70 to 0.73. Clear-sky permanent snow albedos over Antarctica estimated from two independent angular distribution models are consistent to within 0.6%, on average. Despite significant variability in sea ice optical properties with season, the estimated mean relative albedo error is -1 % for very dark sea ice and 0.1% for very bright sea ice when albedos derived from different viewing angles are averaged. The estimated regional root-mean-square (RMS) relative albedo error is 5.6% and 2.6% when the sea ice angular distribution models are applied to a region that contains very dark and very bright sea ice, respectively. Similarly, the estimated relative albedo bias error for fresh snow is -0.1% for very dark snow.

Kato, S.↗

DOE OSTI2022

Graph identification of proteins in tomograms ( GRIP‐Tomo )

Abstract In this study, we present a method of pattern mining based on network theory that enables the identification of protein structures or complexes from synthetic volume densities, without the knowledge of predefined templates or human biases for refinement. We hypothesized that the topological connectivity of protein structures is invariant, and they are distinctive for the purpose of protein identification from distorted data presented in volume densities. Three‐dimensional densities of a protein or a complex from simulated tomographic volumes were transformed into mathematical graphs as observables. We systematically introduced data distortion or defects such as missing fullness of data, the tumbling effect, and the missing wedge effect into the simulated volumes, and varied the distance cutoffs in pixels to capture the varying connectivity between the density cluster centroids in the presence of defects. A similarity score between the graphs from the simulated volumes and the graphs transformed from the physical protein structures in point data was calculated by comparing their network theory order parameters including node degrees, betweenness centrality, and graph densities. By capturing the essential topological features defining the heterogeneous morphologies of a network, we were able to accurately identify proteins and homo‐multimeric complexes from 10 topologically distinctive samples without realistic noise added. Our approach empowers future developments of tomogram processing by providing pattern mining with interpretability, to enable the classification of single‐domain protein native topologies as well as distinct single‐domain proteins from multimeric complexes within noisy volumes.

59 BASIC BIOLOGICAL SCIENCES↗

DOE OSTI2025

Regime Characterization of Offshore Wind Resource Using Unsupervised Learning

Predictability of wind resource conditions is critical for offshore wind design and operations. While many studies of extreme wind conditions focus on specific events such as low-level jets or ramps, these rely on threshold definitions that limit generality. Here we present a data-driven framework that combines principal component analysis (PCA), self-organizing maps (SOM), and k-means clustering to classify wind resource conditions as typical and anomalous from climatological data. Anomalies are defined not by fixed thresholds but by flagging samples located far from SOM node centers inside the baseline SOM structure. This reframes extremes as rare ebents and hence, likely difficult to anticipate by numerical weather prediction models. We applied this approach to 23 years (2000–2022) of hourly profiles from the NOW-23 hindcast model at the Humboldt Wind Energy Area. Classification is conducted on a feature space consisting of 10 m wind speed and direction, bulk shear and veer across 30–270 m, and a low-level jet index. Dimensionality reduction is achieved through PC. A 2 × 3 OM lattice trained on the PCA vectors identified six baseline regimes spanning weak to strong flow states. High quantization-error profiles are identified and re-clustered into four anomalous regimes. The baseline regimes exhibited clear seasonal and diurnal cycles. Meanwhile, the anomalous regimes represented <10 % of all hours but showed distinct combinations of speed, shear, and veer, when compared to the baseline regimes. Anomalous regimes are typically short-lived (~few hours), yet their transitions can lead to hub-height wind changes of −18 to +9 m s -1 . For a representative 15 MW turbine, these shifts imply rapid swings in capacity factor from near-full output to negligible generation. Validation with lidar buoy data showed 51% agreement in SOM labels across ~6,000 overlapping hours, with most mismatches confined to adjacent speed classes. HRRR comparisons further revealed that anomalous regimes were disproportionately associated with forecast biases exceeding 5 m s -1 . Together, these results reframe extremes in offshore wind from absolute maxima or minima to weather states that are difficult to anticipate from models.

17 WIND ENERGY↗

DOE OSTI2023

Subseasonal Representation and Predictability of North American Weather Regimes Using Cluster Analysis

Abstract This study focuses on assessing the representation and predictability of North American weather regimes, which are persistent large-scale atmospheric patterns, in a set of initialized subseasonal reforecasts created using the Community Earth System Model, version 2 (CESM2). The k -means clustering was used to extract four key North American (10°–70°N, 150°–40°W) weather regimes within ERA5 reanalysis, which were used to interpret CESM2 subseasonal forecast performance. Results show that CESM2 can recreate the climatology of the four main North American weather regimes with skill but exhibits biases during later lead times with overoccurrence of the West Coast high regime and underoccurrence of the Greenland high and Alaskan ridge regimes. Overall, the West Coast high and Pacific trough regimes exhibited higher predictability within CESM2, partly related to El Niño. Despite biases, several reforecasts were skillful and exhibited high predictability during later lead times, which could be partly attributed to skillful representation of the atmosphere from the tropics to extratropics upstream of North America. The high predictability at the subseasonal time scale of these case-study examples was manifested as an “ensemble realignment,” in which most ensemble members agreed on a prediction despite ensemble trajectory dispersion during earlier lead times. Weather regimes were also shown to project distinct temperature and precipitation anomalies across North America that largely agree with observational products. This study further demonstrates that unsupervised learning methods can be used to uncover sources and limits of subseasonal predictability, along with systematic biases present in numerical prediction systems. Significance Statement North American weather regimes are large-scale atmospheric patterns that can persist for several days. Their skillful subseasonal (2 weeks or greater) prediction can provide valuable lead time to prepare for temperature and precipitation anomalies that can stress energy and water resources. The purpose of this study was to assess the climatological representation and subseasonal predictability of four key North American weather regimes using a research subseasonal prediction system and clustering analysis. We found that the Pacific trough and West Coast high regimes exhibited higher predictability than other regimes and that skillful representation of conditions across the tropics and extratropics can increase predictability during later lead times. Future work will quantify causal pathways associated with high predictability.

58 GEOSCIENCES↗

NASA NTRSDate not supplied

Extraterrestrial Amino Acids in the C2 Ungrouped Carbonaceous Chondrite Tarda: A Unique Distribution

Introduction: Meteorites provide a record of the chemical processes that occurred in the early solar sys-tem. The delivery of organic matter by carbonaceous chondrites to the early Earth could have been an im-portant source of amino acids and other prebiotic or-ganic molecules required for the emergence of life [1]. To date, 96 amino acids have been named in meteor-ites, most of which are rare or absent in the biosphere [2]. The variability in amino acid concentrations and isomer distributions measured in carbonaceous chon-drites can be explained by differences in parent body chemistry and alteration conditions [3]. In addition, aqueous alteration was likely important in the amplifi-cation of some left-handed amino acids over their right-handed forms (L-excesses up to ~60%) in the Tagish Lake meteorite [3], suggesting that the origin of life on Earth and possibly elsewhere in the solar sys-tem was biased towards L-amino acids. On August 25, 2020, a fireball was witnessed in southern Morocco and the first pieces of the meteorite fall were recovered the following day near the town of Tarda [4]. Thousands of individual fragments were recovered from the strewn field totaling ~4 kg [4]. Based on petrographic observations, bulk mineralogy, and chemical and O-isotopic analyses, Tarda has been classified as a C2 ungrouped carbonaceous chondrite with similarities to Tagish Lake (though more primi-tive), as well as CI, CY, and CR chondrites [4,5]. The fall and rapid recovery of the Tarda stones provide an important opportunity to investigate a C-rich meteorite using the state-of-the-art techniques that will also be used to study the samples returned from asteroids Ryugu and Bennu by the Hayabusa2 and OSIRIS-REx missions, respectively. Here, we report the first amino acid analyses of the Tarda meteorite. The total concentrations, enantio-meric ratios and relative distributions of amino acids were determined using ultrahigh performance liquid chromatography with UV fluorescence and time-of-flight mass spectrometry (LC-FD/ToF-MS) at NASA GSFC. This analytical technique was employed on extracts from two different pre-rain Tarda meteorite fragments and a sample of sand collected from the Tarda fall site for the purpose of assessing potential terrestrial contamination. Bulk H, C, and N elemental and isotopic analyses were also conducted at the Car-negie Institution for Science (CIS) to assist with petro-logic type classification using elemental analysis iso-tope ratio mass spectrometry (EA-IRMS).

D P Glavin↗

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↗