Search NASA⌕ Search

SEARCH · Search NASA

Results for “deep Gaussian processes”

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 91 records · Page 5

MOOSE ProbML: Parallelized probabilistic machine learning and uncertainty quantification for computational energy applications

Here, this paper presents the development and demonstration of massively parallel probabilistic machine learning (ML) and uncertainty quantification (UQ) capabilities within the Multiphysics Object-Oriented Simulation Environment (MOOSE), an open-source computational platform for parallel finite element and finite volume analyses. In addressing the computational expense and uncertainties inherent in complex multiphysics simulations, this paper integrates Gaussian process (GP) variants, active learning, Bayesian inverse UQ, adaptive forward UQ, Bayesian optimization, evolutionary optimization, and Markov chain Monte Carlo (MCMC) within MOOSE. It also elaborates on the interaction among key MOOSE systems — Sampler, MultiApp, Reporter, and Surrogate — in enabling these capabilities. The modularity offered by these systems enables development of a multitude of probabilistic ML and UQ algorithms in MOOSE. Example code demonstrations include parallel active learning and parallel Bayesian inference via active learning. The impact of these developments is illustrated through five applications relevant to computational energy applications: UQ of nuclear fuel fission product release, using parallel active learning Bayesian inference; very rare events analysis in nuclear microreactors using active learning; advanced manufacturing process modeling using multi-output GPs (MOGPs) and dimensionality reduction; fluid flow using deep GPs (DGPs); and tritium transport model parameter optimization for fusion energy, using batch Bayesian optimization. These capabilities are part of the MOOSE framework.

97 - MATHEMATICS AND COMPUTING↗

Modeling I/O Performance Variability Using Conditional Variational Auto Encoders

Storage system performance modeling is crucial for efficient use of heterogeneous shared resources on leadership class computers. Variability in application performance, particularly variability arising from concurrent applications sharing I/O resources, is a major hurdle in the development of accurate performance models. We adopt a deep learning approach based on conditional variational auto encoders (CVAE) for I/O performance modeling, and use it to quantify performance variability. We illustrate our approach using the data collected on Edison, a production supercomputing system at the National Energy Research Scientific Computing Center (NERSC). The CVAE approach is investigated by comparing it to a previously proposed sensitivity-based Gaussian process (GP) model. We find that the CVAE model performs slightly better than the GP model in cases where training and testing data come from different applications, since CVAE can inherently leverage the whole data from multiple applications whereas GP partitions the data and builds separate models for each partition. Hence, the CVAE offers an alternative modeling approach that does not need preprocessing; it has enough flexibility to handle data from a wide variety of applications without changing the inference approach.

I/O perfornance variability↗

The K2-ESPRINT project III: A close-in super-earth around a metal-rich mid-M dwarf

We validate a R{sub p}=2.32±0.24R{sub ⊕} planet on a close-in orbit (P = 2.260455 ± 0.000041 days) around K2-28 (EPIC 206318379), a metal-rich M4-type dwarf in the Campaign 3 field of the K2 mission. Our follow-up observations included multi-band transit observations from the optical to the near-infrared, low-resolution spectroscopy, and high-resolution adaptive optics (AO) imaging. We perform a global fit to all of the observed transits using a Gaussian process-based method and show that the transit depths in all of the passbands adopted for the ground-based transit follow-ups (r{sub 2}{sup ′},z{sub s,2},J,H,K{sub s}) are within ∼2σ of the K2 value. Based on a model of the background stellar population and the absence of nearby sources in our AO imaging, we estimate the probability that a background eclipsing binary could cause a false positive to be <2 × 10{sup −5}. We also show that K2-28 cannot have a physically associated companion of stellar type later than M4, based on the measurement of almost identical transit depths in multiple passbands. There is a low probability for an M4 dwarf companion (≈0.072{sub −0.04}{sup +0.02}), but even if this were the case, the size of K2-28b falls within the planetary regime. K2-28b has the same radius (within 1σ) and experiences irradiation from its host star similar to the well-studied GJ 1214b. Given the relative brightness of K2-28 in the near-infrared (m{sub Kep}=14.85 mag and m{sub H} = 11.03 mag) and relatively deep transit (0.6%–0.7%), a comparison between the atmospheric properties of these two planets with future observations would be especially interesting.

79 ASTRONOMY AND ASTROPHYSICS↗

Combining High-Throughput Experiments and Active Learning to Characterize Deep Eutectic Solvents

The high tunability of deep eutectic solvents (DESs) stems from the ease of changing their precursors and relative compositions. However, measuring the physicochemical properties across large composition and temperature ranges, necessary to properly design target-specific DESs, is tedious and error-prone and represents a bottleneck in the advancement and scalability of DES-based applications. As such, active learning (AL) methodologies based on Gaussian processes (GPs) were developed in this work to minimize the experimental effort necessary to characterize DESs. Owing to its importance for large-scale applications, the reduction of DES viscosity through the addition of a low-molecular-weight solvent was explored as a case study. A high-throughput experimental screening was initially performed on nine different ternary DESs. Then, GPs were successfully trained to predict DES viscosity from its composition and temperature, showcasing the ability of these stochastic, nonparametric models to accurately describe the physicochemical properties of complex mixtures. Finally, the ability of GPs to provide estimates of their own uncertainty was leveraged through an AL framework to minimize the number of data points necessary to obtain accurate viscosity modes. This led to a significant reduction in data requirements, with many systems requiring only five independent viscosity data points to be properly described.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Automatic parameter selection for electron ptychography via Bayesian optimization

Abstract Electron ptychography provides new opportunities to resolve atomic structures with deep sub-angstrom spatial resolution and to study electron-beam sensitive materials with high dose efficiency. In practice, obtaining accurate ptychography images requires simultaneously optimizing multiple parameters that are often selected based on trial-and-error, resulting in low-throughput experiments and preventing wider adoption. Here, we develop an automatic parameter selection framework to circumvent this problem using Bayesian optimization with Gaussian processes. With minimal prior knowledge, the workflow efficiently produces ptychographic reconstructions that are superior to those processed by experienced experts. The method also facilitates better experimental designs by exploring optimized experimental parameters from simulated data.

42 ENGINEERING↗

Exploration with Scalable Gaussian Process Reinforcement Learning

Exploration is a challenging problem in reinforcement learning (RL), especially in environments with sparse rewards. Quantifying and utilizing the parametric uncertainty has been shown to be paramount for successful exploration [Osband et al., 2018]. Bayesian, or approximately Bayesian, methods present a principled means of estimating the parametric uncertainty in RL problems. Gaussian processes, nonparametric Bayesian models, are often impractical due to poor scalability and computational bottlenecks. We introduce a scalable Gaussian process RL (GPRL) method which directly induces sparsity in the covariance matrix to facilitate faster computation. This is a departure from previous GPRL methods which instead rely on data reduction and subsampling. We compare various covariance-based exploration techniques (Thompson sampling, upper confidence bound, and probabilistic maximum variance) which leverage our scalable GP framework in sparse reward environments. Finally, we show favorable comparison against the bootstrapped deep Q-Network.

97 MATHEMATICS AND COMPUTING↗

Strong localization induced anomalous temperature dependence exciton emission above 300 K from SnO{sub 2} quantum dots

SnO{sub 2} quantum dots (QDs) are potential materials for deep ultraviolet (DUV) light emitting devices. In this study, we report the temperature and excitation power-dependent exciton luminescence from SnO{sub 2} QDs. The exciton emission exhibits anomalous blue shift, accompanied with band width reduction with increasing temperature and excitation power above 300 K. The anomalous temperature dependences of the peak energy and band width are well interpreted by the strongly localized carrier thermal hopping process and Gaussian shape of band tails states, respectively. The localized wells and band tails at conduction minimum are considered to be induced by the surface oxygen defects and local potential fluctuation in SnO{sub 2} QDs.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Comparison of the Gaussian Wind Farm Model with Historical Data of Three Offshore Wind Farms

A recent expert elicitation showed that model validation remains one of the largest barriers for commercial wind farm control deployment. The Gaussian-shaped wake deficit model has grown in popularity in wind farm field experiments, yet its validation for larger farms and throughout annual operation remains limited. This article addresses this scientific gap, providing a model comparison of the Gaussian wind farm model with historical data of three offshore wind farms. The energy ratio is used to quantify the model’s accuracy. We assume a fixed turbulence intensity of $I_∞$ = 6% and a standard deviation on the inflow wind direction of $σ_{wd}$ = 3° in our Gaussian model. First, we demonstrate the non-uniqueness issue of $I_∞$ and $σ_{wd}$, which display a waterbed effect when considering the energy ratios. Second, we show excellent agreement between the Gaussian model and historical data for most wind directions in the Offshore Windpark Egmond aan Zee (OWEZ) and Westermost Rough wind farms (36 and 35 wind turbines, respectively) and wind turbines on the outer edges of the Anholt wind farm (110 turbines). Turbines centrally positioned in the Anholt wind farm show larger model discrepancies, likely due to deep-array effects that are not captured in the model. A second source of discrepancy is hypothesized to be inflow heterogeneity. In future work, the Gaussian wind farm model will be adapted to address those weaknesses.

17 WIND ENERGY↗

SELECTION OF BURST-LIKE TRANSIENTS AND STOCHASTIC VARIABLES USING MULTI-BAND IMAGE DIFFERENCING IN THE PAN-STARRS1 MEDIUM-DEEP SURVEY

We present a novel method for the light-curve characterization of Pan-STARRS1 Medium Deep Survey (PS1 MDS) extragalactic sources into stochastic variables (SVs) and burst-like (BL) transients, using multi-band image-differencing time-series data. We select detections in difference images associated with galaxy hosts using a star/galaxy catalog extracted from the deep PS1 MDS stacked images, and adopt a maximum a posteriori formulation to model their difference-flux time-series in four Pan-STARRS1 photometric bands g {sub P1}, r {sub P1}, i {sub P1}, and z {sub P1}. We use three deterministic light-curve models to fit BL transients; a Gaussian, a Gamma distribution, and an analytic supernova (SN) model, and one stochastic light-curve model, the Ornstein-Uhlenbeck process, in order to fit variability that is characteristic of active galactic nuclei (AGNs). We assess the quality of fit of the models band-wise and source-wise, using their estimated leave-out-one cross-validation likelihoods and corrected Akaike information criteria. We then apply a K-means clustering algorithm on these statistics, to determine the source classification in each band. The final source classification is derived as a combination of the individual filter classifications, resulting in two measures of classification quality, from the averages across the photometric filters of (1) the classifications determined from the closest K-means cluster centers, and (2) the square distances from the clustering centers in the K-means clustering spaces. For a verification set of AGNs and SNe, we show that SV and BL occupy distinct regions in the plane constituted by these measures. We use our clustering method to characterize 4361 extragalactic image difference detected sources, in the first 2.5 yr of the PS1 MDS, into 1529 BL, and 2262 SV, with a purity of 95.00% for AGNs, and 90.97% for SN based on our verification sets. We combine our light-curve classifications with their nuclear or off-nuclear host galaxy offsets, to define a robust photometric sample of 1233 AGNs and 812 SNe. With these two samples, we characterize their variability and host galaxy properties, and identify simple photometric priors that would enable their real-time identification in future wide-field synoptic surveys.

79 ASTRONOMY AND ASTROPHYSICS↗

A robust estimator of mutual information for deep learning interpretability

Abstract We develop the use of mutual information (MI), a well-established metric in information theory, to interpret the inner workings of deep learning (DL) models. To accurately estimate MI from a finite number of samples, we present GMM-MI (pronounced ‘Jimmie’), an algorithm based on Gaussian mixture models that can be applied to both discrete and continuous settings. GMM-MI is computationally efficient, robust to the choice of hyperparameters and provides the uncertainty on the MI estimate due to the finite sample size. We extensively validate GMM-MI on toy data for which the ground truth MI is known, comparing its performance against established MI estimators. We then demonstrate the use of our MI estimator in the context of representation learning, working with synthetic data and physical datasets describing highly non-linear processes. We train DL models to encode high-dimensional data within a meaningful compressed (latent) representation, and use GMM-MI to quantify both the level of disentanglement between the latent variables, and their association with relevant physical quantities, thus unlocking the interpretability of the latent representation. We make GMM-MI publicly available in this GitHub repository.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Automated Bayesian high-throughput estimation of plasma temperature and density from emission spectroscopy

Here, this paper introduces a novel approach for automated high-throughput estimation of plasma temperature and density using atomic emission spectroscopy, integrating Bayesian inference with sophisticated physical models. We provide an in-depth examination of Bayesian methods applied to the complexities of plasma diagnostics, supported by a robust framework of physical and measurement models. Our methodology is demonstrated using experimental observations in the field of magneto-inertial fusion, focusing on individual and sequential shot analyses of the Plasma Liner Experiment at LANL. The results demonstrate the effectiveness of our approach in enhancing the accuracy and reliability of plasma parameter estimation and in using the analysis to reveal the deep hidden structure in the data. This study not only offers a new perspective of plasma analysis but also paves the way for further research and applications in nuclear instrumentation and related domains.

Bayesian inference↗

BENCHMARK TESTS FOR MARKOV CHAIN MONTE CARLO FITTING OF EXOPLANET ECLIPSE OBSERVATIONS

Ground-based observations of exoplanet eclipses provide important clues to the planets' atmospheric physics, yet systematics in light curve analyses are not fully understood. It is unknown if measurements suggesting near-infrared flux densities brighter than models predict are real, or artifacts of the analysis processes. We created a large suite of model light curves, using both synthetic and real noise, and tested the common process of light curve modeling and parameter optimization with a Markov Chain Monte Carlo algorithm. With synthetic white noise models, we find that input eclipse signals are generally recovered within 10% accuracy for eclipse depths greater than the noise amplitude, and to smaller depths for higher sampling rates and longer baselines. Red noise models see greater discrepancies between input and measured eclipse signals, often biased in one direction. Finally, we find that in real data, systematic biases result even with a complex model to account for trends, and significant false eclipse signals may appear in a non-Gaussian distribution. To quantify the bias and validate an eclipse measurement, we compare both the planet-hosting star and several of its neighbors to a separately chosen control sample of field stars. Re-examining the Rogers et al. Ks-band measurement of CoRoT-1b finds an eclipse 3190{sup +370}{sub -440} ppm deep centered at {phi}{sub me} = 0.50418{sup +0.00197}{sub -0.00203}. Finally, we provide and recommend the use of selected data sets we generated as a benchmark test for eclipse modeling and analysis routines, and propose criteria to verify eclipse detections.

79 ASTRONOMY AND ASTROPHYSICS↗

Rough surface scattering results based on bandpass autocorrelation forms

Surface-height autocorrelation forms such as Gaussian and exponential are often used in studies of near-normal incidence rough-surface scattering. Such models require the existence of a constant, or DC, value in the spectrum. The consequences of autocorrelation forms that correspond to spectral processes that are essentially bandpass in nature are examined. One such process is that of ocean wind waves. In this case, the spectral components do not extend down to zero frequency. The physical optics backscatter theory is reexamined relative to such autocorrelation functions. Experimental results obtained from a wavetank are compared to the autocorrelation model used in the analysis. The analysis indicates that Gaussian correlation length or mean-square slope is not an appropriate parameter for narrowband conditions and that significant slope is a more relevant parameter. Inherent in the deep-phase assumption is some form of slope dependency. The analysis given (and variants thereof) can be used to provide insight into the physical effects of separate spectral components and of spectral directionality.

Miller, Lee S.↗

A Decision-Making Machine Learning Approach in Hermite Spectral Approximations of Partial Differential Equations

The accuracy and effectiveness of Hermite spectral methods for the numerical discretization of partial differential equations on unbounded domains are strongly affected by the amplitude of the Gaussian weight function employed to describe the approximation space. This is particularly true if the problem is under-resolved, i.e., there are no enough degrees of freedom. The issue becomes even more crucial when the equation under study is time-dependent, forcing in this way the choice of Hermite functions where the corresponding weight depends on time. In order to adapt dynamically the approximation space, it is here proposed an automatic decision-making process that relies on machine learning techniques, such as deep neural networks and support vector machines. The algorithm is numerically tested with success on a simple 1D problem, but the main goal is its exportability in the context of more serious applications. Here we also show at the end an application in the framework of plasma physics.

97 MATHEMATICS AND COMPUTING↗

Effects of correlated noise on the full-spectrum combining and complex-symbol combining arraying techniques

The process of combining telemetry signals received at multiple antennas, commonly referred to as arraying, can be used to improve communication link performance in the Deep Space Network (DSN). By coherently adding telemetry from multiple receiving sites, arraying produces an enhancement in signal-to-noise ratio (SNR) over that achievable with any single antenna in the array. A number of different techniques for arraying have been proposed and their performances analyzed in past literature. These analyses have compared different arraying schemes under the assumption that the signals contain additive white Gaussian noise (AWGN) and that the noise observed at distinct antennas is independent. In situations where an unwanted background body is visible to multiple antennas in the array, however, the assumption of independent noises is no longer applicable. A planet with significant radiation emissions in the frequency band of interest can be one such source of correlated noise. For example, during much of Galileo's tour of Jupiter, the planet will contribute significantly to the total system noise at various ground stations. This article analyzes the effects of correlated noise on two arraying schemes currently being considered for DSN applications: full-spectrum combining (FSC) and complex-symbol combining (CSC). A framework is presented for characterizing the correlated noise based on physical parameters, and the impact of the noise correlation on the array performance is assessed for each scheme.

Vazirani, P.↗

DeepCMB: Lensing reconstruction of the cosmic microwave background with deep neural networks

Next-generation cosmic microwave background (CMB) experiments will have lower noise and therefore increased sensitivity, enabling improved constraints on fundamental physics parameters such as the sum of neutrino masses and the tensor-to-scalar ratio r. Achieving competitive constraints on these parameters requires high signal-to-noise extraction of the projected gravitational potential from the CMB maps. Standard methods for reconstructing the lensing potential employ the quadratic estimator (QE). However, the QE performs suboptimally at the low noise levels expected in upcoming experiments. Other methods, like maximum likelihood estimators (MLE), are under active development. In this work, we demonstrate reconstruction of the CMB lensing potential with deep convolutional neural networks (CNN) - ie, a ResUNet. The network is trained and tested on simulated data, and otherwise has no physical parametrization related to the physical processes of the CMB and gravitational lensing. We show that, over a wide range of angular scales, ResUNets recover the input gravitational potential with a higher signal-to-noise ratio than the QE method, reaching levels comparable to analytic approximations of MLE methods. We demonstrate that the network outputs quantifiably different lensing maps when given input CMB maps generated with different cosmologies. We also show we can use the reconstructed lensing map for cosmological parameter estimation. This application of CNN provides a few innovations at the intersection of cosmology and machine learning. First, while training and regressing on images, we predict a continuous-variable field rather than discrete classes. Second, we are able to establish uncertainty measures for the network output that are analogous to standard methods. We expect this approach to excel in capturing hard-to-model non-Gaussian astrophysical foreground and noise contributions.

79 ASTRONOMY AND ASTROPHYSICS↗

Quantum Simulation of Molecular Dynamics Processes─A Benchmark Study Using a Classical Simulator and Present-Day Quantum Hardware

Here, we explore how the fundamental problems in quantum molecular dynamics can be modeled using classical simulators (emulators) of quantum computers and the actual quantum hardware available to us today. The list of problems we tackle includes propagation of a free wave packet, vibration of a harmonic oscillator, and tunneling through a barrier. Each of these problems starts with the initial wave packet setup. Although Qiskit provides a general method for initializing wave functions, in most cases it generates deep quantum circuits. While these circuits perform well on noiseless simulators, they suffer from excessive noise on quantum hardware. To overcome this issue, we designed a shallower quantum circuit for preparing a Gaussian-like initial wave packet, which improves the performance of real hardware. Next, quantum circuits are implemented to apply the kinetic and potential energy operators for the evolution of a wave function over time. The results of our modeling on classical emulators of quantum hardware agree perfectly with the results obtained using the traditional (classical) methods. This serves as a benchmark and demonstrates that the quantum algorithms and Qiskit codes we developed are accurate. However, the results obtained on the actual quantum hardware available today, such as IBM’s superconducting qubits and IonQ’s trapped ions, indicate large discrepancies due to hardware limitations. This work highlights both the potential and challenges of using quantum computers to solve fundamental quantum molecular dynamics problems.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Enhanced ultraviolet electroluminescence and spectral narrowing from ZnO quantum dots/GaN heterojunction diodes by using high-k HfO{sub 2} electron blocking layer

We demonstrated the capability of realizing enhanced ZnO-related UV emissions by using the low-cost and solution-processable ZnO quantum dots (QDs) with the help of a high-k HfO{sub 2} electron blocking layer (EBL) for the ZnO QDs/p-GaN light-emitting diodes (LEDs). Full-width at half maximum of the LED devices was greatly decreased from ∼110 to ∼54 nm, and recombinations related to nonradiative centers were significantly suppressed with inserting HfO{sub 2} EBL. The electroluminescence of the ZnO QDs/HfO{sub 2}/p-GaN LEDs demonstrated an interesting spectral narrowing effect with increasing HfO{sub 2} thickness. The Gaussian fitting revealed that the great enhancement of the Zn{sub i}-related emission at ∼414 nm whereas the deep suppression of the interfacial recombination at ∼477 nm should be the main reason for the spectral narrowing effect.

36 MATERIALS SCIENCE↗