Search NASA⌕ Search

SEARCH · Search NASA

Results for “random sampling”

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 55 records · Page 3

C-SAW: a framework for graph sampling and random walk on GPUs

Many applications require to learn, mine, analyze and visualize large-scale graphs. These graphs are often too large to be addressed efficiently using conventional graph processing technologies. Fortunately, recent research efforts find out graph sampling and random walk, which significantly reduce the size of original graphs, can benefit the tasks of learning, mining, analyzing and visualizing large graphs by capturing the desirable graph properties. This paper introduces C-SAW, the first framework that accelerates Sampling and Random Walk framework on GPUs. Particularly, C-SAW makes three contributions: First, our framework provides a generic API which allows users to implement a wide range of sampling and random walk algorithms with ease. Second, offloading this framework on GPU, we introduce warp-centric parallel selection, and two novel optimizations for collision migration. Third, towards supporting graphs that exceed the GPU memory capacity, we introduce efficient data transfer optimizations for out-of-memory and multi-GPU sampling, such as workload-aware scheduling and batched multi-instance sampling. Taken together, our framework constantly outperforms the state of the art projects in addition to the capability of supporting a wide range of sampling and random walk algorithms.

97 MATHEMATICS AND COMPUTING↗

Analysis of the Space Propulsion System Problem Using RAVEN

This paper presents the solution of the space propulsion problem using a PRA code currently under development at Idaho National Laboratory (INL). RAVEN (Reactor Analysis and Virtual control ENviroment) is a multi-purpose Probabilistic Risk Assessment (PRA) software framework that allows dispatching different functionalities. It is designed to derive and actuate the control logic required to simulate the plant control system and operator actions (guided procedures) and to perform both Monte- Carlo sampling of random distributed events and Event Tree based analysis. In order to facilitate the input/output handling, a Graphical User Interface (GUI) and a post-processing data-mining module are available. RAVEN allows also to interface with several numerical codes such as RELAP5 and RELAP-7 and ad-hoc system simulators. For the space propulsion system problem, an ad-hoc simulator has been developed and written in python language and then interfaced to RAVEN. Such simulator fully models both deterministic (e.g., system dynamics and interactions between system components) and stochastic behaviors (i.e., failures of components/systems such as distribution lines and thrusters). Stochastic analysis is performed using random sampling based methodologies (i.e., Monte-Carlo). Such analysis is accomplished to determine both the reliability of the space propulsion system and to propagate the uncertainties associated to a specific set of parameters. As also indicated in the scope of the benchmark problem, the results generated by the stochastic analysis are used to generate risk-informed insights such as conditions under witch different strategy can be followed.

97 MATHEMATICS AND COMPUTING↗

Technical Report: Scalable Parallel Algorithms for High Dimensional Numerical Integration

We implemented a scalable parallel quasi-Monte Carlo numerical high-dimensional integration for tera-scale data points. The implemented algorithm uses the Sobol s quasi-sequences to generate random samples. Sobol s sequence was used to avoid clustering effects in the generated random samples and to produce low-discrepancy random samples which cover the entire integration domain. The performance of the algorithm was tested. Obtained results prove the scalability and accuracy of the implemented algorithms. The implemented algorithm could be used in different applications where a huge data volume is generated and numerical integration is required. We suggest using the hyprid MPI and OpenMP programming model to improve the performance of the algorithms. If the mixed model is used, attention should be paid to the scalability and accuracy.

97 MATHEMATICS AND COMPUTING↗

Identifying Important Ions and Positions in Mass Spectrometry Imaging Data Using CUR Matrix Decompositions

Mass spectrometry imaging enables label-free, high-resolution spatial mapping of the chemical composition of complex, biological samples. Typical experiments require selecting ions and/or positions from the images: ions for fragmentation studies to identify keystone compounds and positions for follow up validation measurements using microdissection or other orthogonal techniques. Unfortunately, with modern imaging machines, these must be selected from an overwhelming amount of raw data. Existing techniques to reduce the volume of data, the most popular of which are principle component analysis and non-negative matrix factorization, have the disadvantage that they return difficult-to-interpret linear combinations of actual data elements. In this work, we show that CX and CUR matrix decompositions can be used directly to address this selection need. CX and CUR matrix decompositions use empirical statistical leverage scores of the input data to provide provably good low-rank approximations of the measured data that are expressed in terms of actual ions and actual positions, as opposed to difficult-to-interpret eigenions and eigenpositions. We show that this leads to effective prioritization of information for both ions and positions. In particular, important ions can be found either by using the leverage scores as a ranking function and using a deterministic greedy selection algorithm or by using the leverage scores as an importance sampling distribution and using a random sampling algorithm; however, selection of important positions from the original matrix performed significantly better when they were chosen with the random sampling algorithm. Also, we show that 20 ions or 40 locations can be used to reconstruct the original matrix to a tolerance of 17% error for a widely studied image of brain lipids; and we provide a scalable implementation of this method that is applicable for analysis of the raw data where there are often more than a million rows and/or columns, which is larger than SVD-based low-rank approximation methods can handle. Lastly, these results introduce the concept of CX/CUR matrix factorizations to mass spectrometry imaging, describing their utility and illustrating principled algorithmic approaches to deal with the overwhelming amount of data generated by modern mass spectrometry imaging.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Boundaries of quantum supremacy via random circuit sampling

Abstract Google’s quantum supremacy experiment heralded a transition point where quantum computers can evaluate a computational task, random circuit sampling, faster than classical supercomputers. We examine the constraints on the region of quantum advantage for quantum circuits with a larger number of qubits and gates than experimentally implemented. At near-term gate fidelities, we demonstrate that quantum supremacy is limited to circuits with a qubit count and circuit depth of a few hundred. Larger circuits encounter two distinct boundaries: a return of a classical advantage and practically infeasible quantum runtimes. Decreasing error rates cause the region of a quantum advantage to grow rapidly. At error rates required for early implementations of the surface code, the largest circuit size within the quantum supremacy regime coincides approximately with the smallest circuit size needed to implement error correction. Thus, the boundaries of quantum supremacy may fortuitously coincide with the advent of scalable, error-corrected quantum computing.

97 MATHEMATICS AND COMPUTING↗

Effect of Nonunital Noise on Random-Circuit Sampling

In this work, drawing inspiration from the type of noise present in real hardware, we study the output distribution of random quantum circuits under practical nonunital noise sources with constant noise rates. We show that even in the presence of unital sources such as the depolarizing channel, the distribution, under the combined noise channel, never resembles a maximally entropic distribution at any depth. To show this, we prove that the output distribution of such circuits never anticoncentrates—meaning that it is never too “flat”—regardless of the depth of the circuit. This is in stark contrast to the behavior of noiseless random quantum circuits or those with only unital noise, both of which anticoncentrate at sufficiently large depths. As a consequence, our results shows that the complexity of random-circuit sampling under realistic noise is still an open question, since anticoncentration is a critical property exploited by both state-of-the-art classical hardness and easiness results. Published by the American Physical Society 2024

Physics↗

Simple effective conservative treatment of uncertainty from sparse samples of random functions

This paper examines the variability of predicted responses when multiple stress-strain curves (reflecting variability from replicate material tests) are propagated through a finite element model of a ductile steel can being slowly crushed. Over 140 response quantities of interest (including displacements, stresses, strains, and calculated measures of material damage) are tracked in the simulations. Each response quantity’s behavior varies according to the particular stress-strain curves used for the materials in the model. We desire to estimate response variability when only a few stress-strain curve samples are available from material testing. Propagation of just a few samples will usually result in significantly underestimated response uncertainty relative to propagation of a much larger population that adequately samples the presiding random-function source. A simple classical statistical method, Tolerance Intervals, is tested for effectively treating sparse stress-strain curve data. The method is found to perform well on the highly nonlinear input-to-output response mappings and non-standard response distributions in the can-crush problem. The results and discussion in this paper support a proposition that the method will apply similarly well for other sparsely sampled random variable or function data, whether from experiments or models. Finally, the simple Tolerance Interval method is also demonstrated to be very economical.

36 MATERIALS SCIENCE↗

Organ sample generator for expected treatment dose construction and adaptive inverse planning optimization

Purpose: To create an organ sample generator (OSG) for expected treatment dose construction and adaptive inverse planning optimization. The OSG generates random samples of organs of interest from a distribution obeying the patient specific organ variation probability density function (PDF) during the course of adaptive radiotherapy. Methods: Principle component analysis (PCA) and a time-varying least-squares regression (LSR) method were used on patient specific geometric variations of organs of interest manifested on multiple daily volumetric images obtained during the treatment course. The construction of the OSG includes the determination of eigenvectors of the organ variation using PCA, and the determination of the corresponding coefficients using time-varying LSR. The coefficients can be either random variables or random functions of the elapsed treatment days depending on the characteristics of organ variation as a stationary or a nonstationary random process. The LSR method with time-varying weighting parameters was applied to the precollected daily volumetric images to determine the function form of the coefficients. Eleven h and n cancer patients with 30 daily cone beam CT images each were included in the evaluation of the OSG. The evaluation was performed using a total of 18 organs of interest, including 15 organs at risk and 3 targets. Results: Geometric variations of organs of interest during h and n cancer radiotherapy can be represented using the first 3 {approx} 4 eigenvectors. These eigenvectors were variable during treatment, and need to be updated using new daily images obtained during the treatment course. The OSG generates random samples of organs of interest from the estimated organ variation PDF of the individual. The accuracy of the estimated PDF can be improved recursively using extra daily image feedback during the treatment course. The average deviations in the estimation of the mean and standard deviation of the organ variation PDF for h and n cancer radiotherapy were less than 2 and 1 mm, respectively, for most organs after the second week of treatment. After the first three weeks of treatment, the mean discrepancy of the dose estimation accuracy was within 1% for most of organs, the corresponding standard deviation was within 2.5% for parotids, the brain stem and the cochleae, and within 1% for other organs. Conclusions: A patient specific OSG is feasible and can be used to generate random samples of organs of interest for the expected treatment dose construction and adaptive inverse planning. The accuracy of the OSG can be improved continuously and recursively during the adaptive treatment course using daily volumetric image feedback.

60 APPLIED LIFE SCIENCES↗

Comparison of theory with the experimental characterization of the spatial frequency response of interferometers using a binary pseudo-random array sample

Experimental evaluations of the surface height response of an interference microscope using a binary pseudo-random array test sample are compared with a theory based on a Fourier optics model. Measurements of key instrument characteristics, including the illumination, imaging, and obscuring apertures of three different Mirau objectives, support the theoretical calculations. Agreement between experimental and theoretical modeling confirms the predictability of the spatial frequency response for the purpose of specification and optimization of instrument configuration for specific metrology tasks. The results also provide confidence in methods of compensating for the decrease in instrument response with spatial frequency.

Calibration↗

3013/9975 Surveillance Program Interim Summary Report

The K-Area Materials Storage (KAMS) Documented Safety Analysis (DSA) requires a surveillance program to monitor the safety performance of 3013 containers and 9975 shipping packages stored in KAMS. The SRS surveillance program [Reference 1] outlines activities for field surveillance and laboratory tests that demonstrate the packages meet the functional performance requirements described in the DSA. The SRS program also supports the complexwide Integrated Surveillance Program (ISP) [Reference 2] for 3013 containers. The purpose of this report is to provide a summary of the SRS portion of the surveillance program activities through fiscal year 2010 (FY10) and formally communicate the interpretation of these results by the Surveillance Program Authority (SPA). Surveillance for the initial 3013 container random sampling of the Innocuous bin and the Pressure bin has been completed and there has been no indication of corrosion or significant pressurization. The maximum pressure observed was less than 50 psig, which is well below the design pressure of 699 psig for the 3013 container [Reference 3]. The data collected during surveillance of these bins has been evaluated by the Materials Identification and Surveillance (MIS) Working Group and no additional surveillance is necessary for these bins at least through FY13. A decision will be made whether additional surveillance of these bins is needed during future years of storage and as additional containers are generated. Based on the data collected to date, the SPA concludes that 3013 containers in these bins can continue to be safely stored in KAMS. This year, 13 destructive examinations (DE) were performed on random samples from the Pressure & Corrosion bin. To date, DE has been completed for approximately 30% of the random samples from the Pressure & Corrosion bin. In addition, DE has been performed on 6 engineering judgment (EJ) containers, for a total of 17 to date. This includes one container that exceeded the 3013 Standard moisture limit which was opened at LANL. The container pieces and an oxide sample were sent to SRNL for examination in FY11. No significant pressurization has been observed for the Pressure & Corrosion bin containers. Relatively minor corrosion has been observed on some convenience containers and the inside of two inner containers. While the limited extent of corrosion does not jeopardize the integrity of the outer 3013 containers, it does highlight the importance of continuing to perform DE and the Shelf Life program to assure that the corrosion rate is not accelerating or changing to a different corrosion mechanism (e.g., stress corrosion cracking). Statistical sampling is currently scheduled to be completed in FY17, but there is a proposed reduction of the number of DE's per year for FY11 and beyond which may delay the completion date. Since 3013 containers are stored inside 9975 containers, surveillances of 9975 containers are performed in conjunction with 3013 container surveillances. Results of 9975 container nondestructive examinations (NDEs) and DEs indicate that the containers will provide adequate protection of the 3013 containers in K-Area storage for at least 15 years [Reference 4].

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Investigating the ecological fallacy through sampling distributions constructed from finite populations

Correlation coefficients and linear regression values computed from group averages can differ from correlation coefficients and linear regression values computed using individual scores. This observation known as the ecological fallacy often assumes that all the individual scores are available from a population. In many situations, one must use a sample from the larger population. In such cases, the computed correlation coefficient and linear regression values will depend on the sample that is chosen and the underlying sampling distribution. The sampling distribution of correlation coefficients and linear regression values for group averages will be identical to the sampling distribution for individuals for normally distributed variables for random samples drawn from infinitely large continuous distributions. However, data that is acquired in practice is often acquired when sampling without replacement from a finite population. Our objective is to demonstrate through Monte Carlo simulations that the sampling distributions for correlation and linear regression will also be similar for individuals and group averages when sampling without replacement from normally distributed variables. These simulations suggest that when a random sample from a population is selected, the correlation coefficients and linear regression values computed from individual scores will not be more accurate in estimating the entire population values compared to samples when group averages are used as long as the sample size is the same.

97 MATHEMATICS AND COMPUTING↗

QUANTIFYING KINEMATIC SUBSTRUCTURE IN THE MILKY WAY'S STELLAR HALO

We present and analyze the positions, distances, and radial velocities for over 4000 blue horizontal-branch (BHB) stars in the Milky Way's halo, drawn from SDSS DR8. We search for position-velocity substructure in these data, a signature of the hierarchical assembly of the stellar halo. Using a cumulative 'close pair distribution' as a statistic in the four-dimensional space of sky position, distance, and velocity, we quantify the presence of position-velocity substructure at high statistical significance among the BHB stars: pairs of BHB stars that are close in position on the sky tend to have more similar distances and radial velocities compared to a random sampling of these overall distributions. We make analogous mock observations of 11 numerical halo formation simulations, in which the stellar halo is entirely composed of disrupted satellite debris, and find a level of substructure comparable to that seen in the actually observed BHB star sample. This result quantitatively confirms the hierarchical build-up of the stellar halo through a signature in phase (position-velocity) space. In detail, the structure present in the BHB stars is somewhat less prominent than that seen in most simulated halos, quite possibly because BHB stars represent an older sub-population. BHB stars located beyond 20 kpc from the Galactic center exhibit stronger substructure than at r{sub gc} < 20 kpc.

79 ASTRONOMY AND ASTROPHYSICS↗

In-situ sampling of a large-scale particle simulation for interactive visualization and analysis

We propose storing a random sampling of data from large scale particle simulations, such as the Roadrunner Universe MC{sup 3} cosmological simulation, to be used for interactive post-analysis and visualization. Simulation data generation rates will continue to be far greater than storage bandwidth rates and other limiting technologies by many orders of magnitude. This implies that only a very small fraction of data generated by the simulation can ever be stored and subsequently post-analyzed. The limiting technology in this situation is analogous to the problem in many population surveys: there aren't enough human resources to query a large population. To cope with the lack of resources, statistical sampling techniques are used to create a representative data set of a large population. Mirroring that situation, we propose to store a simulation-time random sampling of the particle data to cope with the bOlllenecks and support interactive, exploratory post-analysis. The particle samples are immediately stored in a level-ol-detail format for post-visualization and analysis, which amortizes the cost of post-processing for interactive visualization. Additionally, we incorporate a system for recording and visualizing sample approximation error information for confidence and importance highlighting.

97 MATHEMATICS AND COMPUTING↗

Bootstrapping outperforms community‐weighted approaches for estimating the shapes of phenotypic distributions

Abstract Estimating phenotypic distributions of populations and communities is central to many questions in ecology and evolution. These distributions can be characterized by their moments (mean, variance, skewness and kurtosis) or diversity metrics (e.g. functional richness). Typically, such moments and metrics are calculated using community‐weighted approaches (e.g. abundance‐weighted mean). We propose an alternative bootstrapping approach that allows flexibility in trait sampling and explicit incorporation of intraspecific variation, and show that this approach significantly improves estimation while allowing us to quantify uncertainty. We assess the performance of different approaches for estimating the moments of trait distributions across various sampling scenarios, taxa and datasets by comparing estimates derived from simulated samples with the true values calculated from full datasets. Simulations differ in sampling intensity (individuals per species), sampling biases (abundance, size), trait data source (local vs. global) and estimation method (two types of community‐weighting, two types of bootstrapping). We introduce the traitstrap R package, which contains a modular and extensible set of bootstrapping and weighted‐averaging functions that use community composition and trait data to estimate the moments of community trait distributions with their uncertainty. Importantly, the first function in the workflow, trait_fill , allows the user to specify hierarchical structures (e.g. plot within site, experiment vs. control, species within genus) to assign trait values to each taxon in each community sample. Across all taxa, simulations and metrics, bootstrapping approaches were more accurate and less biased than community‐weighted approaches. With bootstrapping, a sample size of 9 or more measurements per species per trait generally included the true mean within the 95% CI. It reduced average percent errors by 26%–74% relative to community‐weighting. Random sampling across all species outperformed both size‐ and abundance‐biased sampling. Our results suggest randomly sampling ~9 individuals per sampling unit and species, covering all species in the community and analysing the data using nonparametric bootstrapping generally enable reliable inference on trait distributions, including the central moments, of communities. By providing better estimates of community trait distributions, bootstrapping approaches can improve our ability to link traits to both the processes that generate them and their effects on ecosystems.

Maitner, Brian S.↗

Vehicle energy consumption estimation using large scale simulations and machine learning methods

Vehicle energy consumption and powertrain operations for future vehicle powertrain technologies are predicted through full-vehicle simulations using Autonomie, a simulation tool developed at Argonne National Laboratory. Over 1 million vehicles, with different powertrain types (conventional and numerous electrified powertrains) and component technologies (advanced engines, batteries, electrical machines, light weighting) are simulated for outputs including vehicle fuel economy and cost. Despite the ability of Autonomic to predict fuel economy given various combination of vehicle inputs, the full-vehicle simulation is very time-consuming. It takes approximately 84 h to simulate for 33,060 vehicles even using distributed computing on a cluster of 128 worker nodes. Moreover, Autonomic is unable to populate a continuous output space through simulation since it only yields outputs with discrete values instead of an input-output relation. This paper proposes a novel large-scale learning and prediction process (LSLPP) via machine learning approaches to answer the need for a continuous space of vehicle fuel economy and a more efficient simulation process. By learning in a supervised fashion to discover structures in data, the machine learning approaches seek to train a model that can explain the data with which we can perform prediction over unseen data. As such, we are able to accomplish analytic continuation on the current discrete space of simulation results. In addition, we present a random sampling -based numerosity reduction algorithm incorporated in LSLPP that reduces simulation runs and improves efficiency of data procurement. We analytically derive a theoretical bound in order to guarantee the effectiveness of our algorithm. We also compare random sampling to stratified sampling and suggest sampling strategies. A user-friendly tool is developed to support the process. Our experimental results confirm that the proposed LSLPP is able to greatly accelerate prediction and analysis of vehicle fuel economy. The output of this study is used by the US government to evaluate the impact of its R&D funding on energy security and CO 2 emission as well as for the setting of Corporate Average Fuel Economy (CAFE) standards.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Quantifying uncertainty in uranium concentration measurements via K-edge densitometry

This study quantifies the uncertainty in uranium concentration predictions of fluoride and chloride-based salts within a steel pipe using K-edge densitometry. Modeling and simulation was conducted with the Monte Carlo N-Particle Transport (MCNP) code. The quality of of this technique’s prediction in a pipe requires proper characterization of the pipe’s thickness, which is dependent on the source size and axial offset from the pipe centerline. The thickness was determined as either the center-line thickness seen by the X-ray source or an average value determined through random sampling. Generally, the predicted concentrations were slightly better at lower offset with the random sampling thickness and using the center-line thickness for the highest offsets. For a line-beam source and varying axial offsets, the relative error of concentration was within 1% of the true value but uncertainty increased by 2 orders of magnitude. Similarly, for no axial offset, the relative error was significantly less than 1% while no trend for uncertainty was found. However, at the largest possible offset for a given source size, the concentrations become erroneous and greater than the allowable 1% relative error. Furthermore, high offsets tended to increase the variance of the transmission spectra by 3 orders of magnitude.

Characterization and Analytical Technique↗

Hourglass Sampling of Participants in the Human Reliability Program (HRP) for Drug and Alcohol (D&A) Testing

Hourglass Sampling of Participants in the Human Reliability Program (HRP) for Alcohol and Drug Testing Ivan R. Thomas Idaho National Laboratory The random sampling with replacement of Human Reliability Program (HRP) participants for alcohol and drug testing can have the disadvantage that some participants are selected multiple times while others might not be chosen during an annual testing period. To alleviate this inefficiency, an “hourglass” sampling scheme has been developed at the Idaho National Laboratory for the random selection of HRP participants. With this scheme, all HRP participants are placed in a primary population at the beginning of the calendar year, and throughout the year, sequential random samples (generally of a fixed sample size) are drawn without replacement until the population is emptied. Thus, each participant is guaranteed to be tested at least once annually; but due to the random selection, the time of the initial test is unknown. After initial testing, the participants drawn from the primary population are transferred to a secondary population for potential retesting. Each time that the primary population is sampled, the secondary population is likewise sampled, but the sampling is with replacement. Thus, while the primary population decreases at a constant rate, the secondary population increases at the same rate through the accrual and retention of previously-tested participants, hence the hourglass concept. The replacement sampling of participants from the secondary population is through an increasing sample size (a fixed percentage of those currently in the population). Thus, once in the secondary population, each participant has a constant probability of being reselected, but the number of annual reselections is less than would be realized through traditional replacement sampling from a single population. Furthermore, the objective of maintaining suspense on the part of the HRP participant is retained, that is, all participants, whether or not initially picked from the primary population or reselected from the secondary population, always anticipate a future selection.

98 NUCLEAR DISARMAMENT, SAFEGUARDS, AND PHYSICAL P↗

Iterative self-organizing SCEne-LEvel sampling (ISOSCELES) for large-scale building extraction

Convolutional neural networks (CNN) provide state-of-the-art performance in many computer vision tasks, including those related to remote-sensing image analysis. Successfully training a CNN to generalize well to unseen data, however, requires training on samples that represent the full distribution of variation of both the target classes and their surrounding contexts. With remote sensing data, acquiring a sufficiently representative training set is a challenge due to both the inherent multi-modal variability of satellite or aerial imagery and the general high cost of labeling data. To address this challenge, we have developed ISOSCELES, an Iterative Self-Organizing SCEne LEvel Sampling method for hierarchical sampling of large image sets. Using affinity propagation, ISOSCELES automates the selection of highly representative training images. Compared to random sampling or using available reference data, the distribution of the training is principally data driven, reducing the chance of oversampling uninformative areas or undersampling informative ones. In comparison to manual sample selection by an analyst, ISOSCELES exploits descriptive features, spectral and/or textural, and eliminates human bias in sample selection. Using a hierarchical sampling approach, ISOSCELES can obtain a training set that reflects both between-scene variability, such as in viewing angle and time of day, and within-scene variability at the level of individual training samples. We verify the method by demonstrating its superiority to stratified random sampling in the challenging task of adapting a pre-trained model to a new image and spatial domain for country-scale building extraction. Using a pair of hand-labeled training sets comprising 1,987 sample image chips, a total of 496,000,000 individually labeled pixels, we show, across three distinct model architectures, an increase in accuracy, as measured by F1-score, of 2.2–4.2%.

42 ENGINEERING↗