Search NASA⌕ Search

SEARCH · Search NASA

Results for “statistical learning”

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 1,387 records · Page 77

Validating the RTOG-Endorsed Brachial Plexus Contouring Atlas: An Evaluation of Reproducibility Among Patients Treated by Intensity-Modulated Radiotherapy for Head-and-Neck Cancer

Purpose: To evaluate interobserver variability for contouring the brachial plexus as an organ-at-risk (OAR) and to analyze its potential dosimetric consequences in patients treated with intensity-modulated radiotherapy (IMRT) for head-and-neck cancer. Methods and Materials: Using the Radiation Therapy Oncology Group (RTOG)-endorsed brachial plexus contouring atlas, three radiation oncologists independently delineated the OAR on treatment planning computed-tomography (CT) axial scans from 5 representative patients undergoing IMRT to a prescribed dose of 70 Gy for head-and-neck cancer. Dose-volume histograms for the brachial plexus were calculated, and interobserver differences were quantified by comparing various dosimetric statistics. Qualitative analysis was performed by visually assessing the overlapping contours on a single beam's eye view. Results: Brachial plexus volumes for the 5 patients across observers were 26 cc (18-35 cc), 25 cc (21-30 cc), 29 cc (28-32 cc), 29 cc (23-38 cc), and 29 cc (23-34 cc). On qualitative analysis, minimal variability existed except at the inferolateral portion of the OAR, where slight discrepancies were noted among the physicians. Maximum doses to the brachial plexus ranged from 71.6 to 72.6 Gy, 75.2 to 75.8 Gy, 69.1 to 71.0 Gy, 76.4 to 76.9 Gy, and 70.6 to 71.4 Gy. Respective volumes receiving doses greater than 60 Gy (V60) were 8.6 to 10.9 cc, 6.2 to 8.1 cc, 8.2 to 11.6 cc, 8.3 to 10.5 cc, and 5.6 to 9.8 cc. Conclusion: The RTOG-endorsed brachial plexus atlas provides a consistent set of guidelines for contouring this OAR with essentially no learning curve. Adoption of these contouring guidelines in the clinical setting is encouraged.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Optimizing observables with machine learning for better unfolding

Most measurements in particle and nuclear physics use matrix-based unfolding algorithms to correct for detector effects. In nearly all cases, the observable is defined analogously at the particle and detector level. We point out that while the particle-level observable needs to be physically motivated to link with theory, the detector-level need not be and can be optimized. We show that using deep learning to define detector-level observables has the capability to improve the measurement when combined with standard unfolding methods.

47 OTHER INSTRUMENTATION↗

Comparative study of machine learning techniques for post-combustion carbon capture systems

Computational analysis of countercurrent flows in packed absorption columns, often used in solvent-based post-combustion carbon capture systems (CCSs), is challenging. Typically, computational fluid dynamics (CFD) approaches are used to simulate the interactions between a solvent, gas, and column's packing geometry while accounting for the thermodynamics, kinetics, heat, and mass transfer effects of the absorption process. These simulations can then be used explain a column's hydrodynamic characteristics and evaluate its CO 2 -capture efficiency. However, these approaches are computationally expensive, making it difficult to evaluate numerous designs and operating conditions to improve efficiency at industrial scales. In this work, we comprehensively explore the application of statistical ML methods, convolutional neural networks (CNNs), and graph neural networks (GNNs) to aid and accelerate the scale-up and design optimization of solvent-based post-combustion CCSs. We apply these methods to CFD datasets of countercurrent flows in absorption columns with structured packings characterized by several geometric parameters. We train models to use these parameters, inlet velocity conditions, and other model-specific representations of the column to estimate key determinants of CO 2 -capture efficiency without having to simulate additional CFD datasets. We also evaluate the impact of different input types on the accuracy and generalizability of each model. We discuss the strengths and limitations of each approach to further elucidate the role of CNNs, GNNs, and other machine learning approaches for CO 2 -capture property prediction and design optimization.

97 MATHEMATICS AND COMPUTING↗

Advanced Analytics Studies Applied to US Department of Veterans Affairs' Corporate Data Warehouse (Initial Draft)

In this report, we describe our experience of applying several advanced analytics algorithms to the US Department of Veterans Affairs’ (VA’s) Corporate Data Warehouse (CDW) electronic health records datasets during FY2017-18. While various algorithms were applied to the CDW data, the main goal of this effort was to provide useful insights into the operational aspects (as opposed to the purely clinical aspect) of this specific implementation. Since there are not many reports in the public literature on advanced analytics applied to this data, this report provides unique insight in this regard. We focused on two machine learning (ML) applications: (1) clinical pathway inference (patterns dominating clinical pathways applied to a male dataset of Stable Ischemic Heart Disease patients as use case) and (2) medical-concept representation learning’s capability to inform and refine cohort membership based on probable patient outcomes. During our study on clinical pathway inference, we used two main modeling techniques: (a) topic probabilistic modeling (latent Dirichlet allocation [LDA]) and (b) feature imbedding. We observed that the applicability of LDA to the pathway inference remains in question. LDA provides results that are intuitive and easy for humans to comprehend; however, when applied to the pathway inference, it also generates pathway patterns that consist of a few dominant pathway components. In addition, LDA provides no information regarding the relationships among components in the same pathway. The outputs of LDA are also found to categorize patients based on their trace data of clinical procedures. However, the results suggest that LDA is biased toward statistically dominant components. This makes it particularly hard to discriminate pathway subbranches. To address the issue, we designed a new pathway inference methodology that integrates temporal ordering of pathway components into LDA results, as existing feature embedding tools such as word2vec are not readily applicable to our task. In addition, we developed our own feature embedding tool customized for patient traces. We conclude this study by suggesting two approaches to use embedding representation of a component into LDA and briefly list our lessons learned and recommendations for future work. For our study on medical-concept representation learning ability to inform and refine cohort membership based on probable patient outcomes, during our empirical evaluation we compared methods from medical-concept learning to standard one of a kind (one-of-K) encoding to evaluate the change in effectiveness as done in Choi, Schuetz, Stewart, and Sun (2016). We performed two primary empirical evaluations. The first evaluation was on a curated collection of 60,000 patients with no more than 1 year of medical history included. The second was a collection of patients with no restriction to the amount of medical history included. The hypothesis across both evaluations is that representation learning is broadly useful independent of downstream processing models; thus, for each experiment, we trained three models—a logistic regression model, a two-layer neural network model, and a nearest-neighbor model—and then averaged the results. Each model was trained using fivefold cross validation. We provided evidence that medical-representation learning improves predictions of the primary diagnosis category of a short patient history through 12 experiments. In addition, we provided evidence that medical-representation learning fails to improve prediction of the primary diagnosis category of an arbitrarily long patient history through 12 experiments. We conclude this study with a brief list of our lessons learned and recommendations for future work. We conclude the report with a discussion of recent technology trends as they relate to our artificial intelligence and, more specifically, to our ML research and approaches as described previously.

97 MATHEMATICS AND COMPUTING↗

A Method for Producing Hierarchical and Statistically Calibrated Predictions of Nuclear Material Properties from Existing Models

Computer vision-based analysis of micrographs of nuclear materials is an emerging technique for property prediction, synthetic route identification, and other material analysis tasks. These analysis tasks play a pivotal role in many material characterization applications such as signature development for treaty verification, process optimization, etc. The backbone in many of the recent computer vision-based techniques is a deep learning model, which takes a fixed-size set of pixels and provides a class prediction for that set of pixels. For example, previous work developed a deep convolutional neural network (CNN) to predict the synthetic route from a 256 px x 256 px patch taken from a larger image of uranium ore concentrates. In this work, we present several methods for first calibrating these models in a manner that they can provide accurate probabilities of their predictions’ veracity, and several methods of combining these probabilities. Overall, the combination of these two steps into a pipeline allows for full-image and even full-sample (where a sample has many images) predictions with associated confidence values. Finally, we show that one can also use the patch predictions and confidence to produce a visualization to map predicted constituents through the image. Results and examples for predicting and mapping uranium ore concentrates’ synthetic process from imagery will be presented.

artificial intelligence↗

Extracting forces from noisy dynamics in dusty plasmas

Extracting environmental forces from noisy data is a common yet challenging task in complex physical systems. Machine learning (ML) represents a robust approach to this problem, yet is mostly tested on simulated data with known parameters. Here we use supervised ML to extract the electrostatic, dissipative, and stochastic forces acting on micron-sized charged particles levitated in an argon plasma (dusty plasma). By tracking the sub-pixel motion of particles in subsequent images, we successfully estimated these forces from their random motion. The experiments contained important sources of non-Gaussian noise, such as drift and pixel-locking, representing a data mismatch from methods used to analyze simulated data with purely Gaussian noise. Our model was trained on simulated particle trajectories that included all of these artifacts, and used more than 100 dynamical and statistical features, resulting in a prediction with 50\% better accuracy than conventional methods. Lastly, in systems with two interacting particles, the model provided non-contact measurements of the particle charge and Debye length in the plasma environment.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

WE-D-BRE-05: Prediction of Late Radiation-Induced Proctitis in Prostate Cancer Patients Using Chromosome Aberration and Cell Proliferation Rate

Purpose: Chromosome damage and cell proliferation rate have been investigated as potential biomarkers for the early prediction of late radiationinduced toxicity. Incorporating these endpoints, we explored the predictive power for late radiation proctitis using a machine learning method. Methods: Recently, Beaton et al. showed that chromosome aberration and cell proliferation rate could be used as biomarkers to predict late radiation proctitis (Beaton et al. (2013) Int J Rad Onc Biol Phys, 85:1346–1352). For the identification of radiosensitive biomarkers, blood samples were collected from 10 patients with grade 3 late proctitis along with 20 control patients with grade 0 proctitis. After irradiation at 6 Gy, statistically significant difference was observed between the two groups, using the number of dicentrics and excess fragments, and the number of cells in metaphase 2 (M2). However, Beaton et al. did not show the usefulness of combining these endpoints. We reanalyzed the dataset to investigate whether incorporating these endpoints can increase the predictive power of radiation proctitis, using a support vector machine (SVM). Results: Using the SVM method with the number of fragments and M2 endpoints, perfect classification was achieved. In addition, to avoid biased estimate of the classification method, leave-one-out cross-validation (LOO-CV) was performed. The best performance was achieved when all three endpoints were used with 87% accuracy, 90% sensitivity, 85% specificity, and 0.85 AUC (the area under the receiver operating characteristic (ROC) curve). The most significant endpoint was the number of fragments that obtained 83% accuracy, 70% sensitivity, 90% specificity, and 0.82 AUC. Conclusion: We demonstrated that chromosome damage and cell proliferation rate could be significant biomarkers to predict late radiation proctitis. When these endpoints were used together in conjunction with a machine learning method, the better performance was obtained. However, evaluation for other late radiation toxicities is necessary to elucidate these observations.

60 APPLIED LIFE SCIENCES↗

Classification Analytics of Pu-239 and U-235 Source Signatures Using Gamma Spectral Regions

Machine learning detection methods using gamma signatures from spectral measurements of low-intensity Pu-239 and U-235 sources are studied. NaI detectors located at different distances fromthe source have been used to collect the training and independent testing data sets. The source is introduced via a shielded conduit into the facility where it is surrounded by 21 NaI detectors deployed over 6 x 6 meters area in the formation of two concentric circles and a spiral. The counts in gamma spectral regions associated with these two sources are estimated at 1 second intervals for each NaI detector, and are used as classifier features for detecting the source presence. Eight different classifiers with five basic properties — namely, smooth, non-smooth, statistical, structural, and hyper-parameter tuning — are trained and tested using the background and source measurements collected over multiple experimental runs. While the overall classifier performance improved as detectors closer to the source are used, some identically produced detectors under-performed but differently between two sources. Some classifiers achieved lower training error but their testing error based on independent measurements is higher for both sources. Overall, these results indicate significant over-fitting by these methods, and illustrate the complexity of training and selecting the machine learning methods to solve these detection problems.

Rao, Nageswara↗

Applications of Autonomous Data Collection and Active Learning

Advances in sensors and robotics have dramatically improved the diversity of experimental approaches available to the materials community. Autonomous data collection platforms, either custom-made or commercially available, provide researchers with novel tools with which to probe materials behavior and perform advanced materials characterization. The application of novel control algorithms and active learning approaches can create much more robust experimental data, or can be used to improve the performance of existing characterization tools. Five papers within this special topic focus on experimental and computational methodologies for use in automatic data collection routines for materials characterization. From novel platforms for materials discovery to new statistical frameworks for assessing the autonomous experimentation process, these five papers highlight the diverse range of applications of automation for advancing materials science.

36 MATERIALS SCIENCE↗

Quantum learning without quantum memory

A quantum learning machine for binary classification of qubit states that does not require quantum memory is introduced and shown to perform with the minimum error rate allowed by quantum mechanics for any size of the training set. This result is shown to be robust under (an arbitrary amount of) noise and under (statistical) variations in the composition of the training set, provided it is large enough. This machine can be used an arbitrary number of times without retraining. Its required classical memory grows only logarithmically with the number of training qubits, while its excess risk decreases as the inverse of this number, and twice as fast as the excess risk of an ‘‘estimate-and-discriminate’’ machine, which estimates the states of the training qubits and classifies the data qubit with a discrimination protocol tailored to the obtained estimates.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

HIERARCHICAL PROBABILISTIC INFERENCE OF COSMIC SHEAR

Point estimators for the shearing of galaxy images induced by gravitational lensing involve a complex inverse problem in the presence of noise, pixelization, and model uncertainties. We present a probabilistic forward modeling approach to gravitational lensing inference that has the potential to mitigate the biased inferences in most common point estimators and is practical for upcoming lensing surveys. The first part of our statistical framework requires specification of a likelihood function for the pixel data in an imaging survey given parameterized models for the galaxies in the images. We derive the lensing shear posterior by marginalizing over all intrinsic galaxy properties that contribute to the pixel data (i.e., not limited to galaxy ellipticities) and learn the distributions for the intrinsic galaxy properties via hierarchical inference with a suitably flexible conditional probabilitiy distribution specification. We use importance sampling to separate the modeling of small imaging areas from the global shear inference, thereby rendering our algorithm computationally tractable for large surveys. With simple numerical examples we demonstrate the improvements in accuracy from our importance sampling approach, as well as the significance of the conditional distribution specification for the intrinsic galaxy properties when the data are generated from an unknown number of distinct galaxy populations with different morphological characteristics.

79 ASTRONOMY AND ASTROPHYSICS↗

Towards an optimal flow: Density-of-states-informed replica-exchange simulations

Here we learn that replica exchange (RE) is one of the most popular enhanced-sampling simulations technique in use today. Despite widespread successes, RE simulations can sometimes fail to converge in practical amounts of time, e.g., when sampling around phase transitions, or when a few hard-to-find configurations dominate the statistical averages. We introduce a generalized RE scheme, density-of-states-informed RE, that addresses some of these challenges. The key feature of our approach is to inform the simulation with readily available, but commonly unused, information on the density of states of the system as the RE simulation proceeds. This enables two improvements, namely, the introduction of resampling moves that actively move the system towards equilibrium and the continual adaptation of the optimal temperature set. As a consequence of these two innovations, we show that the configuration flow in temperature space is optimized and that the overall convergence of RE simulations can be dramatically accelerated.

97 MATHEMATICS AND COMPUTING↗

Baycal

Bayesian Model Calibration (BayCal) toolkit is a software plugin for Risk Analysis Virtual Environment (RAVEN) framework, arming at inversely quantifying the uncertainties associated with simulation model parameters based on available experiment data. BayCal seeks statistical inference of the uncertain input parameters that are consistent with the available measurement data or observed data. The unique feature of BayCal is the capability to be linked with RAVEN to build corresponding calibration workflows for complex multi-physics simulations. In addition to be able to use the machine learning capability of RAVEN to significantly reduce the computational cost of expensive simulation models, another distinctive feature of BayCal is the capability to deal with high-dimensional correlated model outputs, such as time series observations at multiple locations, via principal component analysis (PCA) technique.

Wang, Congjian↗

Field Trial of a Low-Cost, Distributed Plug Load Monitoring System

Researchers have struggled to inventory and characterize the energy use profiles of the ever-growing category of so-called miscellaneous electric loads (MELs) because plug-load monitoring is cost-prohibitive to the researcher and intrusive to the homeowner. However, these data represent a crucial missing link to understanding how homes use energy. Detailed energy use profiles would enable the nascent automated home energy management (AHEM) industry to develop effective control algorithms that target consumer electronics and other plug loads. If utility and other efficiency programs are to incent AHEM devices, they need large-scale datasets that provide statistically meaningful justification of their investments by quantifying the aggregate energy savings achievable. To address this need, NREL researchers investigated a variety of plug-load measuring devices available commercially and tested them in the laboratory to identify the most promising candidates for field applications. This report centers around the lessons learned from a field validation of one proof-of-concept system, called Smartenit (formerly SimpleHomeNet). The system was evaluated based on the rate of successful data queries, reliability over a period of days to weeks, and accuracy. This system offers good overall performance when deployed with up to 10 end nodes in a residential environment, although deployment with more nodes and in a commercial environment is much less robust. NREL concludes that the current system is useful in selected field research projects, with the recommendation that system behavior is observed over time.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Inversion of surface parameters using fast learning neural networks

A neural network approach to the inversion of surface scattering parameters is presented. Simulated data sets based on a surface scattering model are used so that the data may be viewed as taken from a completely known randomly rough surface. The fast learning (FL) neural network and a multilayer perceptron (MLP) trained with backpropagation learning (BP network) are tested on the simulated backscattering data. The RMS error of training the FL network is found to be less than one half the error of the BP network while requiring one to two orders of magnitude less CPU time. When applied to inversion of parameters from a statistically rough surface, the FL method is successful at recovering the surface permittivity, the surface correlation length, and the RMS surface height in less time and with less error than the BP network. Further applications of the FL neural network to the inversion of parameters from backscatter measurements of an inhomogeneous layer above a half space are shown.

Dawson, M. S.↗

Motion sickness and proprioceptive aftereffects following virtual environment exposure

To study the potential aftereffects of virtual environments (VE), tests of visually guided behavior and felt limb position (pointing with eyes open and closed) along with self-reports of motion sickness-like discomfort were administered before and after 30 min exposure of 34 subjects. When post- discomfort was compared to a pre-baseline, the participants reported more sickness afterward (p < 0.03). The change in felt limb position resulted in subjects pointing higher (p < 0.038) and slightly to the left, although the latter difference was not statistically significant (p = 0.08). When findings from a second study using a different VE system were compared, they essentially replicated the results of the first study with higher sickness afterward (p < 0.001) and post- pointing errors were also up (p < 0.001) and to the left (p < 0.001). While alternative explanations (e.g. learning, fatigue, boredom, habituation, etc.) of these outcomes cannot be ruled out, the consistency of the post- effects on felt limb position changes in the two VE implies that these recalibrations may linger once interaction with the VE has concluded, rendering users potentially physiologically maladapted for the real world when they return. This suggests there may be safety concerns following VE exposures until pre-exposure functioning has been regained. The results of this study emphasize the need for developing and using objective measures of post-VE exposure aftereffects in order to systematically determine under what conditions these effects may occur.

NASA Discipline Neuroscience↗

Classification of Ascension Island and Natal Ozonesondes Using Self-Organizing Maps

Ozone profiles from balloon-borne ozonesondes are used for development of satellite algorithms and in chemistry-climate model initialization, assimilation and evaluation. An important issue in the application of these profiles is how best to treat variations where varying photochemical and dynamical influences can cause the ozone mixing ratio in the tropospheric segments of the profile to change by of a factor of 2-3 within a day. Clustering techniques are an ideal way to approach the statistical classification of profile data and we apply self-organizing maps to tropical tropospheric SHADOZ data, hypothesizing that the data will sort according to various influences on ozone, namely anthropogenic sources like biomass burning, meteorological conditions, and stratospheric or extra-tropical intrusions. Self-organizing maps, that use a learning algorithm to reveal the most prominent features of a data set according to a specified number of clusters, have been determined for the 1998-2009 SHADOZ profiles over Ascension Island (512 profiles, 7.98 deg. S, 14.42 deg. W) and Natal, Brazil (425 profiles, 5.42degS, 35.38degW). The 2 × 2 self-organizing map, which creates 4 clusters, reveals that deviations from the average ozone in the free troposphere include both increased ozone resulting from seasonal biomass burning in Africa and locally reduced ozone brought about by convective lifting of unpolluted boundary-layer air. Expanding to a 4 × 4 self-organizing map shows how biomass burning influences the yearly cycle of tropospheric ozone at Ascension Island and captures the seasonality of ozone at both Ascension Island and Natal. Comparing Ascension Island and Natal using a 4 × 4 self-organizing map at each site reveals similarities in mid-tropospheric ozone, but shows differences in lower-tropospheric ozone due to Ascension Island being closer to African biomass burning and more affected by descent from the mean Walker circulation, with less convective activity, than Natal.

algorithms↗

Analysis and Prediction of Weather Impacted Ground Stop Operations

When the air traffic demand is expected to exceed the available airport's capacity for a short period of time, Ground Stop (GS) operations are implemented by Federal Aviation Administration (FAA) Traffic Flow Management (TFM). The GS requires departing aircraft meeting specific criteria to remain on the ground to achieve reduced demands at the constrained destination airport until the end of the GS. This paper provides a high-level overview of the statistical distributions as well as causal factors for the GSs at the major airports in the United States. The GS's character, the weather impact on GSs, GS variations with delays, and the interaction between GSs and Ground Delay Programs (GDPs) at Newark Liberty International Airport (EWR) are investigated. The machine learning methods are used to generate classification models that map the historical airport weather forecast, schedule traffic, and other airport conditions to implemented GS/GDP operations and the models are evaluated using the cross-validations. This modeling approach produced promising results as it yielded an 85% overall classification accuracy to distinguish the implemented GS days from the normal days without GS and GDP operations and a 71% accuracy to differentiate the GS and GDP implemented days from the GDP only days.

Analysis↗