Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Neural Networks (Computer)”

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 541 records · Page 30

Big Data Synchrophasor Monitoring and Analytics for Resiliency Tracking (BDSMART)

This report contains key findings from a project titled Big Data Synchrophasor Monitoring and Analytics for Resiliency Tracking (BDSMART), which was carried out through a collaborative effort of a team of researchers from Texas A&M Engineering Experiment Station, Temple University, and Quanta Technology, LLC. The in-kind support came from OSIsoft (acquired by AVEVA), which provided their PI Historian software to demonstrate the use case of streaming PMU data. The first section of the report describes the project goals and objectives related to the development of Machine Learning (ML) models capable of detecting and classifying events by processing phasor measurements captured in the field by Phasor Measurement Units (PMUs). The data for this study was contributed by the utilities/ISOs from the Western and Eastern interconnects and ERCOT, further referred to as Interconnect B (IC B), Interconnect A (IC A), and Interconnect C (IC C), respectively. The approach that the BDSMART Research Team proposed and the key research tasks defined by the team are outlined in this section. The next section describes the technical approach. We first discuss the data constraints related to the PMU measurements and data interpretation constraints imposed by the data contributors. They provided neither the topological information of the grid nor PMU placement locations and captured recorded data at very few locations in the system with the reporting rate of either 30 or 60 fps. The recordings are mostly positive sequence voltage, frequency, and ROCOF, and in some limited cases, three-phase voltages and currents. We then reflect on the bad data issues that stem from poor recording practices and vague definitions of the PMU status bits to supposedly be used for bad data identification. Finally, the data discovery points to imprecise time stamps with incomplete event start/end time, as well as inconsistent and incomplete event labeling, which combined make the implementation of the data models using supervising learning quite challenging. Following the data discovery study, we hypothesize that because the IC B data has the most complete labels, we should focus our model development on that data and then test it on data from other interconnects. We also define the common metrics used to evaluate the results from the ML algorithm tests. We concluded this section by summarizing the common ML models we used and explaining how we implemented and tested them. The issues from this section are expanded in the Training Dataset Report from this project. The final section of this report deals with the accomplishments and conclusions. As the accomplishments, we formulate the problem we are solving and what is achieved by solving the problem. We then reflect on each of the analytics tools we developed and point out the performance of each tool when applied to solving the mentioned problems. We reference this work for further details to the papers we published on each tool. In the conclusions, we give recommendations on how to improve future PMU recording practices to facilitate the ML algorithm implementation and guidance for the future standardization work aimed at clarifying the ambiguities associated with the PMU status bits. We finally list future tasks that can bring about further improvements in the proposed algorithms. The issues from this section are expanded in the Training, and Test Dataset Report filed at the project completion date.

97 MATHEMATICS AND COMPUTING↗

Optimizing Performance on Trinity Utilizing Machine Learning, Proxy Applications and Scheduling Priorities

The sheer number of nodes continues to increase in today’s supercomputers, the first half of Trinity alone contains more than 9400 compute nodes. Since the speed of today’s clusters are limited by the slowest nodes, it more important than ever to identify slow nodes, improve their performance if it can be done, and assure minimal usage of slower nodes during performance critical runs. This is an ongoing maintenance task that occurs on a regular basis and, therefore, it is important to minimize the impact upon its users by assessing and addressing slow performing nodes and mitigating their consequences while minimizing down time. These issues can be solved, in large part, through a systematic application of fast running hardware assessment tests, the application of Machine Learning, and making use of performance data to increase efficiency of large clusters. Proxy applications utilizing both MPI and OpenMP were developed to produce data as a substitute for long runtime applications to evaluate node performance. Machine learning is applied to identify underperforming nodes, and policies are being discussed to both minimize the impact of underperforming nodes and increase the efficiency of the system. In this paper, I will describe the process used to produce quickly performing proxy tests, consider various methods to isolate the outliers, and produce ordered lists for use in scheduling to accomplish this task.

97 MATHEMATICS AND COMPUTING↗

Data supporting manuscript from L. Sheneman, G. Stephanopoulos, A.E. Vasdekis titled "Deep learning classification of lipid droplets in quantitative phase images" as currently under review at PLOS ONE. This includes: 1) raw and binary labeled Quantitative Phase Images (QPI) of Y. lipolytica cells used in the analyses described within the manuscript. 2) various derived data including classifier scores, etc.

Data supporting manuscript from L. Sheneman, G. Stephanopoulos, A.E. Vasdekis titled "Deep learning classification of lipid droplets in quantitative phase images" as currently under review at PLOS ONE. This includes: 1) raw and binary labeled Quantitative Phase Images (QPI) of Y. lipolytica cells used in the analyses described within the manuscript. 2) various derived data including classifier scores, etc.

ANN↗

Algorithms to Improve Training for Deep Learning with Diabetic Retinopathy Images

This is a minisymposium presentation I plan to give at the SIAM conference on Computational Science and Engineering on March 1st, 2021. The work is based on research from my SDRD this year and describes a new pooling method we've developed called variable stride. The goal of the minisymposium is to highlight some of the ground-breaking work being done across the National Nuclear Security Administration by graduate students and Postdocs.

97 MATHEMATICS AND COMPUTING↗

A geometric framework for momentum-based optimizers for low-rank training

Low-rank pre-training and fine-tuning have recently emerged as promising techniques for reducing the computational and storage costs of large neural networks. Training low-rank parameterizations typically relies on conventional optimizers such as heavy ball momentum methods or Adam. In this work, we identify and analyze potential difficulties that these training methods encounter when used to train low-rank parameterizations of weights. In particular, we show that classical momentum methods can struggle to converge to a local optimum due to the geometry of the underlying optimization landscape. To address this, we introduce novel training strategies derived from dynamical low-rank approximation, which explicitly account for the underlying geometric structure. Our approach leverages and combines tools from dynamical low-rank approximation and momentum-based optimization to design optimizers that respect the intrinsic geometry of the parameter space. We validate our methods through numerical experiments, demonstrating faster convergence, and stronger validation metrics at given parameter budgets.

Schotthoefer, Steffen [ORNL] (ORCID:00000002156965↗

Neuromorphic Photonics Based on Phase Change Materials

Neuromorphic photonics devices based on phase change materials (PCMs) and silicon photonics technology have emerged as promising solutions for addressing the limitations of traditional spiking neural networks in terms of scalability, response delay, and energy consumption. In this review, we provide a comprehensive analysis of various PCMs used in neuromorphic devices, comparing their optical properties and discussing their applications. We explore materials such as GST (Ge 2 Sb 2 Te 5 ), GeTe-Sb 2 Te 3 , GSST (Ge 2 Sb 2 Se 4 Te 1 ), Sb 2 S 3 /Sb 2 Se 3 , Sc 0.2 Sb 2 Te 3 (SST), and In 2 Se 3 , highlighting their advantages and challenges in terms of erasure power consumption, response rate, material lifetime, and on-chip insertion loss. By investigating the integration of different PCMs with silicon-based optoelectronics, this review aims to identify potential breakthroughs in computational performance and scalability of photonic spiking neural networks. Further research and development are essential to optimize these materials and overcome their limitations, paving the way for more efficient and high-performance photonic neuromorphic devices in artificial intelligence and high-performance computing applications.

36 MATERIALS SCIENCE↗

Neural Network-Enhanced Reproducing Kernel Particle Method for Image-Based Multiphysics Damage Modeling of Energy Storage Materials

Energy storage materials undergo significant stresses during charge/discharge cycling, which makes understanding their reliability and durability fundamental in predicting performance and service life. Strong electrochemical-mechanical coupling and highly anisotropic material properties contribute to the formation and propagation of micro-cracking, largely along material interfaces and grain boundaries. With microstructural images supplied by the National Renewable Energy Laboratory (NREL), image-based modeling techniques are used to represent the complex material microstructures that dictate the coupled physics of these systems. Traditional electrochemical-mechanical models rely on mesh-based finite element methods, which can lead to difficulties in capturing crack propagation due to mesh dependency. Additionally, commonly used damage models, such as the continuous damage model and the cohesive zone model, often have steep tradeoffs between discontinuous field accuracy and computational expense. In this work, a neural network-enhanced reproducing kernel particle method (NN-RKPM) [1] is leveraged to accurately capture damage and crack propagation throughout the material by learning the location, orientation, and sharpness of discontinuity while allowing for a coarser nodal distribution than that necessary for capturing sharp solution transitions using traditional mesh-based methods. NN-RKPM is used to inform how crack opening and closure in turn affect the coupled chemical equations and material microstructure. Reference: [1] Baek, J., Chen, J. S., Susuki, K., "Neural Network enhanced Reproducing Kernel Particle Method for Modeling Localizations," International Journal for Numerical Methods in Engineering, Vol. 123, pp 4422-4454, https://doi.org/10.1002/nme.7040, 2022.

damage modeling↗

Bias-Variance Trade-Off in Physics-Informed Neural Networks with Randomized Smoothing for High-Dimensional PDEs

Physics-Informed Neural Networks (PINNs) have triggered a paradigm shift in scientific computing, leveraging mesh-free properties and robust approximation capabilities. While proving effective for low-dimensional partial differential equations (PDEs), the computational cost of PINNs remains a hurdle in high-dimensional scenarios. This is particularly pronounced when computing high-order and high-dimensional derivatives in the physics-informed loss. Randomized Smoothing PINN (RS-PINN) introduces Gaussian noise for stochastic smoothing of the original neural net model, enabling the use of Monte Carlo methods for derivative approximation, which eliminates the need for costly automatic differentiation. Despite its computational efficiency, especially in the approximation of high-dimensional derivatives, RS-PINN introduces biases in both loss and gradients, negatively impacting convergence, especially when coupled with stochastic gradient descent (SGD) algorithms. We present a comprehensive analysis of biases in RS-PINN, attributing them to the nonlinearity of the Mean Squared Error (MSE) loss as well as the intrinsic nonlinearity of the PDE itself. We propose tailored bias correction techniques, delineating their application based on the order of PDE nonlinearity. The derivation of an unbiased RS-PINN allows for a detailed examination of its advantages and disadvantages compared to the biased version. Specifically, the biased version has a lower variance and runs faster than the unbiased version, but it is less accurate due to the bias. To optimize the bias-variance trade-off, we combine the two approaches in a hybrid method that balances the rapid convergence of the biased version with the high accuracy of the unbiased version. In addition to methodological contributions, we present an enhanced implementation of RS-PINN. Extensive experiments on diverse high-dimensional PDEs, including Fokker-Planck, Hamilton-Jacobi-Bellman (HJB), viscous Burgers’, Allen-Cahn, and Sine-Gordon equations, illustrate the bias-variance trade-off and highlight the effectiveness of the hybrid RS-PINN. Empirical guidelines are provided for selecting biased, unbiased, or hybrid versions, depending on the dimensionality and nonlinearity of the specific PDE problem.

97 MATHEMATICS AND COMPUTING↗

Mean Field Theory of Self‐Organizing Memristive Connectomes

Abstract Biological neuronal networks are characterized by nonlinear interactions and complex connectivity. Given the growing impetus to build neuromorphic computers, understanding physical devices that exhibit structures and functionalities similar to biological neural networks is an important step toward this goal. Self‐organizing circuits of nanodevices are at the forefront of the research in neuromorphic computing, as their behavior mimics synaptic plasticity features of biological neuronal circuits. However, an effective theory to describe their behavior is lacking. This study provides for the first time an effective mean field theory for the emergent voltage‐induced polymorphism of circuits of a nanowire connectome, showing that the behavior of these circuits can be explained by a low‐dimensional dynamical equation. The equation can be derived from the microscopic dynamics of a single memristive junction in analytical form. The effective model is tested on experiments of nanowire networks and show that it fits both the potentiation and depression of these synapse‐mimicking circuits. It is shown that this theory applies beyond the case of nanowire networks by formulating a general mean‐field theory of conductance transitions in self‐organizing memristive connectomes.

Caravelli, Francesco↗

Deep neural network improves the estimation of polygenic risk scores for breast cancer

Polygenic risk scores (PRS) estimate the genetic risk of an individual for a complex disease based on many genetic variants across the whole genome. Here, we compared a series of computational models for estimation of breast cancer PRS. A deep neural network (DNN) was found to outperform alternative machine learning techniques and established statistical algorithms, including BLUP, BayesA, and LDpred. In the test cohort with 50% prevalence, the Area Under the receiver operating characteristic Curve (AUC) were 67.4% for DNN, 64.2% for BLUP, 64.5% for BayesA, and 62.4% for LDpred. BLUP, BayesA, and LPpred all generated PRS that followed a normal distribution in the case population. However, the PRS generated by DNN in the case population followed a bimodal distribution composed of two normal distributions with distinctly different means. This suggests that DNN was able to separate the case population into a high-genetic-risk case subpopulation with an average PRS significantly higher than the control population and a normal-genetic-risk case subpopulation with an average PRS similar to the control population. This allowed DNN to achieve 18.8% recall at 90% precision in the test cohort with 50% prevalence, which can be extrapolated to 65.4% recall at 20% precision in a general population with 12% prevalence. Interpretation of the DNN model identified salient variants that were assigned insignificant p values by association studies, but were important for DNN prediction. These variants may be associated with the phenotype through nonlinear relationships.

59 BASIC BIOLOGICAL SCIENCES↗

Neural Phase Simulation

The Neural Phase Simulation (NPS) is a package of codes for simulating microstructure evolution and accelerated molecular dynamics with deep neural-networks based surrogate models. NPS is designed to offer quantitatively accurate and computationally efficient simulation capabilities by leveraging modern machine-learning techniques. The primary intended use cases of NPS are training neural network surrogate models, though performing simulations on a single node is also supported. The NPS surrogate models can be trained from ground truth simulation methods, which are supposed to be accurate but expensive, such as molecular dynamics, phase field methods, kinetic Monte Carlo and discrete dislocation dynamics.

Zhou, Fei↗

Emerging Jets Search, Triton Server Deployment, and Track Quality Development: Machine Learning Applications in High Energy Physics

Machine learning is becoming prevalent in high energy physics, with numerous applications in physics analyses and event reconstruction showing great improvements compared to traditional computing methods. This thesis studies three projects which each propose new avenues for machine learning applications within the high energy physics CMS experiment located at CERN. In the first project, a search for a dark matter signal called “emerging jets” is performed, using graph neural networks to greatly increase sensitivity to the signal’s signature within the data. The result of this dark matter search sets the most stringent exclusion limits to date on theoretical emerging jet models. Motivated by inefficiencies encountered when processing the emerging jet graph neural network at Fermi National Accelerator Laboratory’s computing centers, the second project re-optimizes the computing centers for machine learning inference. This re-optimization uses NVIDIA Triton Inference Servers to process users’ analysis code heterogeneously, therefore achieving high processing throughput and decreasing user time-to-insight. The last project focuses on an upgrade to the CMS experiment’s real-time event selection system which improves physics object reconstruction under harsh processing conditions. A boosted decision tree is used to quickly and efficiently quantify a reconstructed particle’s “track quality” in order to remove particle tracks reconstructed erroneously. In summary, this thesis will not only present examples of how high energy physics can greatly benefit by leveraging machine learning techniques for physics analysis and reconstruction, but will also provide guidance on how the field can prepare for the inevitable increase in machine learning applications.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Muon Neutrino Disappearance in NOvA with a Deep Convolutional Neural Network Classifier

The NuMI Off-axis Neutrino Appearance Experiment (NOvA) is designed to study neutrino oscillation in the NuMI (Neutrinos at the Main Injector) beam. NOvA observes neutrino oscillation using two detectors separated by a baseline of 810 km; a 14 kt Far Detector in Ash River, MN and a functionally identical 0.3 kt Near Detector at Fermilab. The experiment aims to provide new measurements of Δm 2 and θ23 and has potential to determine the neutrino mass hierarchy as well as observe CP violation in the neutrino sector. Essential to these analyses is the classification of neutrino interaction events in NOvA detectors. Raw detector output from NOvA is interpretable as a pair of images which provide orthogonal views of particle interactions. A recent advance in the field of computer vision is the advent of convolutional neural networks, which have delivered top results in the latest image recognition contests. This work presents an approach novel to particle physics analysis in which a convolutional neural network is used for classification of particle interactions. The approach has been demonstrated to improve the signal efficiency and purity of the event selection, and thus physics sensitivity. Early NOvA data has been analyzed (2.74×10 20 POT, 14 kt equivalent) to provide new best- fit measurements of sin 2 (θ23) = 0.43 (with a statistically-degenerate compliment near 0.60) and |Δm2 | = 2.48 × 10 -3 eV 2 .

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Robust and scalable uncertainty estimation with conformal prediction for machine-learned interatomic potentials

Abstract Uncertainty quantification (UQ) is important to machine learning (ML) force fields to assess the level of confidence during prediction, as ML models are not inherently physical and can therefore yield catastrophically incorrect predictions. Established a-posteriori UQ methods, including ensemble methods, the dropout method, the delta method, and various heuristic distance metrics, have limitations such as being computationally challenging for large models due to model re-training. In addition, the uncertainty estimates are often not rigorously calibrated. In this work, we propose combining the distribution-free UQ method, known as conformal prediction (CP), with the distances in the neural network’s latent space to estimate the uncertainty of energies predicted by neural network force fields. We evaluate this method (CP+latent) along with other UQ methods on two essential aspects, calibration, and sharpness, and find this method to be both calibrated and sharp under the assumption of independent and identically-distributed (i.i.d.) data. We show that the method is relatively insensitive to hyperparameters selected, and test the limitations of the method when the i.i.d. assumption is violated. Finally, we demonstrate that this method can be readily applied to trained neural network force fields with traditional and graph neural network architectures to obtain estimates of uncertainty with low computational costs on a training dataset of 1 million images to showcase its scalability and portability. Incorporating the CP method with latent distances offers a calibrated, sharp and efficient strategy to estimate the uncertainty of neural network force fields. In addition, the CP approach can also function as a promising strategy for calibrating uncertainty estimated by other approaches.

74 ATOMIC AND MOLECULAR PHYSICS↗

Distilling the Essential Elements of Nuclear Binding via Neural-Network Quantum States

To distill the essential elements of nuclear binding, we seek the simplest Hamiltonian capable of modeling atomic nuclei with percent-level accuracy. A critical aspect of this endeavor consists of accurately solving the quantum many-body problem without incurring an exponential computing cost with the number of nucleons. Furthermore, we address this challenge by leveraging a variational Monte Carlo method based on a highly expressive neural-network quantum state ansatz. In addition to computing binding energies and charge radii of nuclei with up to 𝐴 = 20 nucleons, by evaluating their magnetic moments, we demonstrate that neural-network quantum states are able to correctly capture the self-emerging nuclear shell structure. To this end, we introduce a novel computational protocol based on adding an external magnetic field to the nuclear Hamiltonian, which allows the neural network to learn the preferred polarization of the nucleus within the given magnetic field.

Binding energy & masses↗

Structural inference of networked dynamical systems with universal differential equations

Networked dynamical systems are common throughout science in engineering; e.g., biological networks, reaction networks, power systems, and the like. For many such systems, nonlinearity drives populations of identical (or near-identical) units to exhibit a wide range of nontrivial behaviors, such as the emergence of coherent structures (e.g., waves and patterns) or otherwise notable dynamics (e.g., synchrony and chaos). Here, we seek to infer (i) the intrinsic physics of a base unit of a population, (ii) the underlying graphical structure shared between units, and (iii) the coupling physics of a given networked dynamical system given observations of nodal states. These tasks are formulated around the notion of the Universal Differential Equation, whereby unknown dynamical systems can be approximated with neural networks, mathematical terms known a priori (albeit with unknown parameterizations), or combinations of the two. We demonstrate the value of these inference tasks by investigating not only future state predictions but also the inference of system behavior on varied network topologies. The effectiveness and utility of these methods are shown with their application to canonical networked nonlinear coupled oscillators.

97 MATHEMATICS AND COMPUTING↗