Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “supervised training”

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 469 records · Page 26

Reverse-mode differentiation in arbitrary tensor network format: with application to supervised learning.

This paper describes an efficient reverse-mode differentiation algorithm for contraction operations of tensor networks that may have arbitrary and unconventional network topologies. The approach leverages the tensor contraction tree of Evenbly and Pfeifer (2014), which provides an instruction set for the contraction sequence of a network. We show that this tree can be efficiently leveraged for differentiation of a full tensor network contraction using a recursive scheme that exploits (1) the bilinear property of contraction and (2) the property that trees have single path from root to leaves. While differentiation of tensor-tensor contraction is already possible in most automatic differentiation packages, we show that exploiting these two additional properties in the specific context of contraction sequences can improve efficiency. Following a description of the algorithm and computational complexity analysis, we investigate its utility for gradient-based supervised learning for low-rank function recovery and for fitting real-world unstructured datasets. We demonstrate improved performance over alternating least-squares optimization approaches and the capability to handle heterogeneous and arbitrary tensor network formats. When compared to alternating minimization algorithms, we find that the gradient-based approach requires a smaller oversampling ratio (number of samples compared to number model parameters) for recovery. This increased efficiency extends to fitting unstructured data of varying dimensionality and when employing a variety of tensor network formats. Here, we show improved learning using the hierarchical Tucker method over the tensor-train in high-dimensional settings on a number of benchmark problems.

97 MATHEMATICS AND COMPUTING↗

An anomaly detection framework for dynamic systems using a Bayesian hierarchical framework

Complex systems are susceptible to many types of anomalies, faults, and abnormal behavior caused by a variety of off-nominal conditions that may ultimately result in major failures or catastrophic events. Early and accurate detection of these anomalies using system inputs and outputs collected from sensors and smart devices has become a challenging problem and an active area of research in many application domains. In this article, we present a new Bayesian hierarchical framework that is able to model the relationship between system inputs (sensor measurements) and outputs (response variables) without imposing strong distributional/parametric assumptions while using only a subset of training samples and sensor attributes. Then, an optimal cost-sensitive anomaly detection framework is proposed to determine whether a sample is an anomalous one taking into consideration the trade-off between misclassification errors and detection rates. The model can be used for both supervised and unsupervised settings depending on the availability of data regarding the behavior of the system under anomaly conditions. The unsupervised model is particularly useful when it is prohibitive to identify in advance the anomalies that a system may present and where no data are available regarding the behavior of the system under anomaly conditions. A Bayesian hierarchical setting is used to structure the proposed framework and help with accommodating uncertainty, imposing interpretability, and controlling the sparsity and complexity of the proposed anomaly detection framework. A Markov chain Monte Carlo algorithm is also developed for model training using past data. Furthermore, the numerical experiments conducted using a simulated data set and a wind turbine data set demonstrate the successful application of the proposed work for system response modeling and anomaly detection.

17 WIND ENERGY↗

SU-D-18C-06: Initial Experience with Implementing MRI Safety Guidelines for Patients with Pacemakers - Medical Physicist Perspective

Purpose: Several institutions have developed MRI guidelines for patients with MR-unsafe or MR-conditional pacemakers. Here we highlight the role of a medical physicist in implementing these guidelines for non-pacemaker dependent patients. Guidelines: Implementing these guidelines requires involvement from several medical specialties and a strong collaboration with the site MRI supervisor to develop a structured workflow. A medical physicist is required to be present during the scan to supervise the MR scanning and to maintain a safety checklist that ensures: 1) uninterrupted patient communication with the technologist, 2) continuous patient physiologic monitoring (e.g. blood pressure and electrocardiography) by a trained nurse, 3) redundant patient vitals monitoring (e.g. pulse oximetry) due to the possibility of in vivo electrocardiography reading fluctuations during image acquisition. A radiologist is strongly recommended to be available to review the images before patients are discharged from the scanner. Pacemaker MRI should be restricted to 1.5T field strength. The MRI sequences should be optimized by the physicist with regards to: a) SAR: limited to <1.5 W/Kg for MR-unsafe pacemakers in normal operating mode, b) RF exposure time: <30 min, c) Coils: use T/R coils but not restricted to such, d) Artifacts: further optimization of sequences whenever image quality is compromised due to the pacemaker. In particular, cardiac, breast and left-shoulder MRIs are most susceptible to these artifacts. Possible strategies to lower the SAR include: a) BW reduction, 2) echo-train-length reduction, 3) increase TR, 4) decrease number of averages, 5) decrease flip angle, 6) reduce slices and/or a combination of all the options. Conclusion: A medical physicist in collaboration with the MR supervisor plays an important role in the supervision/implementation of safe MR scanning of pacemaker patients. Developing and establishing a workflow has enabled our institution to scan over 30 patients with pacemakers without complications, including 3 cardiac MR exams.

60 APPLIED LIFE SCIENCES↗

Mitochondrial Morphological Features Are Associated with Fission and Fusion Events

Mitochondria are dynamic organelles that undergo constant remodeling through the regulation of two opposing processes, mitochondrial fission and fusion. Although several key regulators and physiological stimuli have been identified to control mitochondrial fission and fusion, the role of mitochondrial morphology in the two processes remains to be determined. To address this knowledge gap, we investigated whether morphological features extracted from time-lapse live-cell images of mitochondria could be used to predict mitochondrial fate. That is, we asked if we could predict whether a mitochondrion is likely to participate in a fission or fusion event based on its current shape and local environment. Using live-cell microscopy, image analysis software, and supervised machine learning, we characterized mitochondrial dynamics with single-organelle resolution to identify features of mitochondria that are predictive of fission and fusion events. A random forest (RF) model was trained to correctly classify mitochondria poised for either fission or fusion based on a series of morphological and positional features for each organelle. Of the features we evaluated, mitochondrial perimeter positively correlated with mitochondria about to undergo a fission event. Similarly mitochondrial solidity (compact shape) positively correlated with mitochondria about to undergo a fusion event. Our results indicate that fission and fusion are positively correlated with mitochondrial morphological features; and therefore, mitochondrial fission and fusion may be influenced by the mechanical properties of mitochondrial membranes.

59 BASIC BIOLOGICAL SCIENCES↗

Criticality Safety Evaluation of the LLNL Inherently Safe Subcritical Assembly (ISSA)

The LLNL Nuclear Criticality Safety Division has developed a training center to illustrate criticality safety and reactor physics concepts through hands-on experimental training. The experimental assembly, the Inherently Safe Subcritical Assembly (ISSA), uses surplus highly enriched research reactor fuel configured in a water tank. The training activities will be conducted by LLNL following the requirements of an Integration Work Sheet (IWS) and associated Safety Plan. Students will be allowed to handle the fissile material under the supervision of LLNL instructors. This report provides the technical criticality safety basis for instructional operations with the ISSA experimental assembly.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

The Bird project: Using Big Data tools to support Search Analytics

The Bird project explored the use of big data analytics tool to improve the findability of information within the Sandia internal network. We were able to perform query classification utilizing the supervised learning algorithms in the Apache Spark library. By relying on the distributed processing capabilities provided by the Apache Hadoop framework, we successfully processed the large query log files needed to train the models in this effort. The capabilities developed in this project are being used to enhance the effectiveness of the enterprise search engine.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗

Mapping Impervious Surface Expansion using Medium-resolution Satellite Image Time Series: A Case Study in the Yangtze River Delta, China

Cities have been expanding rapidly worldwide, especially over the past few decades. Mapping the dynamic expansion of impervious surface in both space and time is essential for an improved understanding of the urbanization process, land-cover and land-use change, and their impacts on the environment. Landsat and other medium-resolution satellites provide the necessary spatial details and temporal frequency for mapping impervious surface expansion over the past four decades. Since the US Geological Survey opened the historical record of the Landsat image archive for free access in 2008, the decades-old bottleneck of data limitation has gone. Remote-sensing scientists are now rich with data, and the challenge is how to make best use of this precious resource. In this article, we develop an efficient algorithm to map the continuous expansion of impervious surface using a time series of four decades of medium-resolution satellite images. The algorithm is based on a supervised classification of the time-series image stack using a decision tree. Each imerpervious class represents urbanization starting in a different image. The algorithm also allows us to remove inconsistent training samples because impervious expansion is not reversible during the study period. The objective is to extract a time series of complete and consistent impervious surface maps from a corresponding times series of images collected from multiple sensors, and with a minimal amount of image preprocessing effort. The approach was tested in the lower Yangtze River Delta region, one of the fastest urban growth areas in China. Results from nearly four decades of medium-resolution satellite data from the Landsat Multispectral Scanner (MSS), Thematic Mapper (TM), Enhanced Thematic Mapper plus (ETM+) and China-Brazil Earth Resources Satellite (CBERS) show a consistent urbanization process that is consistent with economic development plans and policies. The time-series impervious spatial extent maps derived from this study agree well with an existing urban extent polygon data set that was previously developed independently. The overall mapping accuracy was estimated at about 92.5% with 3% commission error and 12% omission error for the impervious type from all images regardless of image quality and initial spatial resolution.

Gao, Feng↗

Deep Learning Prediction of Interspecies Interactions from Self-organized Spatiotemporal Patterns of Co-evolving Organisms

Microorganisms colonizing natural habits such as soils co-evolve to form specific spatial patterns through interspecies interactions. These self-organized patterns are a key ecological phenotype, which provides critical information on their interaction mechanisms. However, conventional network inference techniques that analyze species population data in bulk have yet to be extended to account for such spatial heterogeneity. Here we proposed supervised deep learning as a new network inference tool for predicting interspecies interactions from spatiotemporal patterns of microbial evolution. Due to lack of biological imaging data that can be used for training deep learning networks, we used in silico data generated from high-fidelity agent-based models to determine model structure and parameters. Even though networks were trained under simple configurations where interaction coefficients are assumed to be spatially invariant, we demonstrated that the resulting model can be utilized to successfully predict spatial variation of interactions in more complex domains (i.e., configured with a context-dependent mixture of interaction coefficients) as well as in simple domains without further training. In the further test against real biological data obtained through imaging experiments of a binary consortium (Pseudomonas fluorescens and a mutant of Escherichia coli), our model also predicted the dramatic shifts in interactions of the two organisms across different environmental contexts. Through various successful demonstrations in this work, the combined use of the agent-based model and machine learning algorithm provides a means to use new type of data - microscopic images - for extracting microbial interactions, therefore presenting itself as a useful tool for the analysis of more complex microbial community interactions.

Lee, Joon-Yong↗

Unsupervised Clustering and Supervised Regression Learning to Select High Temperature Oxidation-Resistant Materials

High temperature oxidation and corrosion degradation mechanisms dictate the lifetime of materials critical to energy production. The combination of modeling and experimental approaches such as machine learning (ML) and data analytics, with sufficient experimental data, can accelerate the development of new materials while limiting its cost. In the present work, ML will be applied to two high temperature oxidation data libraries (Oak Ridge National Laboratory and National Air and Space Administration) that comprised of about 5000 mass change sample datasheets for a variety of materials and temperatures in dry air and air + 10 % H2O. A python code was developed to prepare the data for machine learning by collecting and formatting oxidation rate constants, alloy compositions and environment of exposure into a single data frame. Scikit-learn library and Statistics and Machine Learning Toolbox within MathWorks were then used to perform unsupervised clustering and supervised regression learning. The impact of dataset distribution on the performance of the developed ML models was evaluated. Potential strategies to improve the predictions and enhance extrapolative capability of the previously trained model were investigated.

Romedenne, Marie [ORNL] (ORCID:0000000317936561)↗

Interactive volumetric segmentation for textile micro-tomography data using wavelets and nonlocal means

This work addresses segmentation of volumetric images of woven carbon fiber textiles from micro-tomography data. We propose a semi-supervised algorithm to classify carbon fibers that requires sparse input as opposed to completely labeled images. The main contributions are: (a) design of effective discriminative classifiers, for three-dimensional textile samples, trained on wavelet features for segmentation; (b) coupling of previous step with nonlocal means as simple, efficient alternative to the Potts model; and (c) demonstration of reuse of classifier to diverse samples containing similar content. We evaluate our work by curating test sets of voxels in the absence of a complete ground truth mask. The algorithm obtains an average 0.95 F1 score on test sets and average F1 score of 0.93 on new samples. Finally, we conclude with discussion of failure cases and propose future directions toward analysis of spatiotemporal high-resolution micro-tomography images.

3D image processing↗

Quantum anomaly detection for collider physics

We explore the use of Quantum Machine Learning (QML) for anomaly detection at the Large Hadron Collider (LHC). In particular, we explore a semi-supervised approach in the four-lepton final state where simulations are reliable enough for a direct background prediction. This is a representative task where classification needs to be performed using small training datasets - a regime that has been suggested for a quantum advantage. We find that Classical Machine Learning (CML) benchmarks outperform standard QML algorithms and are able to automatically identify the presence of anomalous events injected into otherwise background-only datasets.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Augmenting Graph Convolution with Distance Preserving Embedding for Improved Learning

Graph convolution incorporates topological information of a graph into learning. Message passing corresponds to traversal of a local neighborhood in classical graph algorithms. We show that incorporating additional global structures, such as shortest paths, through distance preserving embedding can improve performance. Our approach, Gavotte, significantly improves the performance of a range of popular graph neu-ral networks such as GCN, GA T,Graph SAGE, and GCNII for transductive learning. Gavotte also improves the performance of graph neural networks for full-supervised tasks, albeit to a smaller degree. As high-quality embeddings are generated by Gavotte as a by-product, we leverage clustering algorithms on these embed dings to augment the training set and introduce Gavotte+. Our results of Gavotte+ on datasets with very few labels demonstrate the advantage of augmenting graph convolution with distance preserving embedding.

Cong, Guojing↗

Korean coastal water depth/sediment and land cover mapping /1:25,000/ by computer analysis of Landsat imagery

Computer analysis was applied to single data Landsat MSS imagery of a coastal area near Seoul, Korea equivalent to a 1:50,000 topographic map, and featuring large dynamic sediment transport processes. Supervised image processing yielded a test classification map containing five water depth/sediment classes, two shoreline/tidal classes and five coastal land cover classes at a scale of 1:25,000 and with a training set accuracy of 76%; the training sets were selected by direct examination of the digitally displayed imagery. The unsupervised ISOCLAS (Senkus, 1976) clustering analysis was performed to assess the relative value of this approach to image classification in areas of sparse or nonexistent ground control. Results indicate that it is feasible to produce quantitative maps for detailed study of dynamic coastal processes given a Landsat image data base at sufficiently frequent time intervals.

Park, K. Y.↗

Image-to-Image Wildfire Detection via Quantum-Compatible Variational Segmentation from Remotely-sensed Data

Over the last decade, the incidence of wildfires has surged, causing widespread destruction globally. To better comprehend and manage these incidents, remote sensing and aerial missions have been implemented in recent efforts. However, this has resulted in an exponential rise in the amount of remote sensing data utilization, leading to a need for intelligent automation of data extraction in wildfire studies. Machine learning provides an accurate automated approach for detecting these natural anomalies and facilitates decision-makers to take prompt actions. To make insightful decisions in wildfire management, it is imperative to move beyond simple detection and explore the potential of probabilistic generative machine learning for creating "what-if" scenarios for various wildfire conditions. Such models offer improved representation of the stochastic nature of wildfire events. However, the optimization of these models can be computationally expensive, especially when using classical computers. Quantum computers have recently emerged as a promising solution to reduce the computational cost of training such models and improve their performance. In this study, we aim to utilize quantum-compatible machine learning techniques to implement our probabilistic generative approach. To that end, we propose a supervised probabilistic variational model consisting of a U-NET-based image-to-image component along with encoder and decoder networks which work as a variational autoencoder (VAE) component. Additionally, we explore the type of latent distribution type in the VAE component and implement different means for modeling the prior distribution. We further investigate the quantum-compatible versions of the model compared to the classical counterpart and benchmark potential benefits of quantum compatibility over the classical model.

quantum machine learning↗

ChemNet: A Transferable and Generalizable Deep Neural Network for Small-Molecule Property Prediction

With access to large datasets, deep neural networks through representation learning have been able to identify patterns from raw data, achieving human-level accuracy in image and speech recognition tasks. However, in chemistry, availability of large standardized and labelled datasets is scarce, and with a multitude of chemical properties of interest, chemical data is inherently small and fragmented. In this work, we explore transfer learning techniques in conjunction with the existing Chemception CNN model, to create a transferable and generalizable deep neural network for small-molecule property prediction. Our latest model, ChemNet learns in a semi-supervised manner from inexpensive labels computed from the ChEMBL database. When fine-tuned to the Tox21, HIV and FreeSolv dataset, which are 3 separate chemical tasks that ChemNet was not originally trained on, we demonstrate that ChemNet exceeds the performance of existing Chemception models, contemporary MLP models that trains on molecular fingerprints, and it matches the performance of the ConvGraph algorithm, the current state-of-the-art. Furthermore, as ChemNet has been pre-trained on a large diverse chemical database, it can be used as a universal “plug-and-play” deep neural network, which accelerates the deployment of deep neural networks for the prediction of novel small-molecule chemical properties.

Artificial Intelligence↗

Bayesian Safety Risk Modeling of Human-Flightdeck Automation Interaction

Usage of automatic systems in airliners has increased fuel efficiency, added extra capabilities, enhanced safety and reliability, as well as provide improved passenger comfort since its introduction in the late 80's. However, original automation benefits, including reduced flight crew workload, human errors or training requirements, were not achieved as originally expected. Instead, automation introduced new failure modes, redistributed, and sometimes increased workload, brought in new cognitive and attention demands, and increased training requirements. Modern airliners have numerous flight modes, providing more flexibility (and inherently more complexity) to the flight crew. However, the price to pay for the increased flexibility is the need for increased mode awareness, as well as the need to supervise, understand, and predict automated system behavior. Also, over-reliance on automation is linked to manual flight skill degradation and complacency in commercial pilots. As a result, recent accidents involving human errors are often caused by the interactions between humans and the automated systems (e.g., the breakdown in man-machine coordination), deteriorated manual flying skills, and/or loss of situational awareness due to heavy dependence on automated systems. This paper describes the development of the increased complexity and reliance on automation baseline model, named FLAP for FLightdeck Automation Problems. The model development process starts with a comprehensive literature review followed by the construction of a framework comprised of high-level causal factors leading to an automation-related flight anomaly. The framework was then converted into a Bayesian Belief Network (BBN) using the Hugin Software v7.8. The effects of automation on flight crew are incorporated into the model, including flight skill degradation, increased cognitive demand and training requirements along with their interactions. Besides flight crew deficiencies, automation system failures and anomalies of avionic systems are also incorporated. The resultant model helps simulate the emergence of automation-related issues in today's modern airliners from a top-down, generalized approach, which serves as a platform to evaluate NASA developed technologies

Ancel, Ersin↗

Optimization and supervised machine learning methods for fitting numerical physics models without derivatives

Here, we address the calibration of a computationally expensive nuclear physics model for which derivative information with respect to the fit parameters is not readily available. Of particular interest is the performance of optimization-based training algorithms when dozens, rather than millions or more, of training data are available and when the expense of the model places limitations on the number of concurrent model evaluations that can be performed. As a case study, we consider the Fayans energy density functional model, which has characteristics similar to many model fitting and calibration problems in nuclear physics. We analyze hyperparameter tuning considerations and variability associated with stochastic optimization algorithms and illustrate considerations for tuning in different computational settings.

97 MATHEMATICS AND COMPUTING↗

PreMevE Update: Forecasting Ultra-Relativistic Electrons Inside Earth's Outer Radiation Belt

Energetic electrons inside Earth's Van Allen belts pose a major radiation threat to spaceborne electronics that often play vital roles in modern society. Ultra-relativistic electrons with energies greater than or equal to two megaelectron-volt (MeV) are of particular interest, and thus forecasting these ≥2 MeV electrons has a significant meaning to all space sectors. Here, we update the latest development of the predictive model for MeV electrons in the outer radiation belt. The new version, called PREdictive MEV Electron (PreMevE)-2E, forecasts ultra-relativistic electron flux distributions across the outer belt, with no need for in situ measurements of the trapped MeV electron population except at the geosynchronous orbit (GEO). Model inputs include precipitating electrons observed in low-Earthorbits by NOAA satellites, upstream solar wind speeds and densities from solar wind monitors, as well as ultra-relativistic electrons measured by one Los Alamos GEO satellite. We evaluated 32 supervised machine learning models that fall into four different classes of linear and neural network architectures, and successfully tested ensemble forecasting by using groups of top-performing models. All models are individually trained, validated, and tested by in situ electron data from NASA's Van Allen Probes mission. It is shown that the final ensemble model outperforms individual models at most L-shells, and this PreMevE-2E model can provide 25-h (~1-day) and 50-h (~2-day) forecasts with high mean performance efficiency and correlation values. Our results also suggest that this new model is dominated by nonlinear components at L-shells <~4 for ultra-relativistic electrons, different from the dominance of linear components for 1 MeV electrons as previously discovered.

79 ASTRONOMY AND ASTROPHYSICS↗