Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “bayesian optimization”

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 757 records · Page 42

Jensen–Shannon divergence based novel loss functions for Bayesian neural networks

Bayesian neural networks (BNNs) are state-of-the-art machine learning methods that can naturally regularize and systematically quantify uncertainties using their stochastic parameters. Kullback–Leibler (KL) divergence-based variational inference used in BNNs suffer from unstable optimization and challenges in approximating light-tailed posteriors due to the unbounded nature of the KL divergence. To resolve these issues, we formulate a novel loss function for BNNs based on a new modification to the generalized Jensen–Shannon (JS) divergence, which is bounded. In addition, we propose a Geometric JS divergence-based loss, which is computationally efficient since it can be evaluated analytically. We found that the JS divergence-based variational inference is intractable, and hence employed a constrained optimization framework to formulate these losses. Our theoretical analysis and empirical experiments on multiple regression and classification data sets suggest that the proposed losses perform better than the KL divergence-based loss, especially when the data sets are noisy or biased. Specifically, there are approximately 5% and 8% improvements in accuracy for a noise-added CIFAR-10 dataset and a regression dataset, respectively. There is about 13% reduction in false negative predictions of a biased histopathology dataset. Additionally, we quantify and compare the uncertainty metrics for the regression and classification tasks.

97 MATHEMATICS AND COMPUTING↗

Optimizing the Performance of Radionuclide Identification Software in the Hunt for Nuclear Security Threats

The Radionuclide Analysis Kit (RNAK), my team’s most recent nuclide identification software, is entering the testing phase. A question arises: will removing rare nuclides from the software’s library improve its overall performance? An affirmative response indicates fundamental errors in the software’s framework, while a negative response confirms the effectiveness of the software’s key machine learning algorithms. After thorough testing, I found that the performance of RNAK cannot be improved with the library choice effect, thus verifying the effectiveness of RNAK’s algorithms—multiple linear regression, Bayesian network using the Viterbi algorithm, and branch and bound search.

97 MATHEMATICS AND COMPUTING↗

Application of Neutron Multiplicity Counting Experiments to Optimal Cross-Section Adjustments

This paper presents the first application of model calibration to neutron multiplicity counting (NMC) experiments for cross-section optimization that is informed by adjoint-based sensitivity analysis (SA) and first-order uncertainty quantification (UQ). We summarize previous work on SA applied to NMC and describe notable modifications and additions. We give the procedure for first-order UQ and Bayesian-inference-based parameter estimation (PE). We then discuss model calibration applied to NMC of a 4.5-kg sphere of weapons-grade, alpha-phase plutonium metal (the BeRP ball) with the nPod neutron multiplicity counter. For the BeRP ball in bare and polyethylene-reflected configurations, we discuss the sensitivity of the first- and second-moment detector responses (i.e., first and second moments of the NMC distribution, respectively) to the cross sections. We describe the sources of uncertainty in the measured and simulated responses. Specifically, uncertainty in the measured responses is due to both random and systematic sources. Uncertainty in the simulated responses is due to the cross-section covariances. We describe in detail the adjustment to the cross sections and cross-section covariances due to the optimization. Due to the contribution of systematic uncertainties to the measured response uncertainties, the adjustment to the cross sections is similar in trend but larger in magnitude compared to that recommended by previous work. We compare the measured responses to responses simulated with nominal and optimized cross sections, demonstrating that the best-estimate cross sections produce simulations of NMC experiments that are more accurate with reduced uncertainty.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Radar-Based Bayesian Estimation of Ice Crystal Growth Parameters within a Microphysical Model

The potential for polarimetric Doppler radar measurements to improve predictions of ice microphysical processes within an idealized model–observational framework is examined. In an effort to more rigorously constrain ice growth processes (e.g., vapor deposition) with observations of natural clouds, a novel framework is developed to compare simulated and observed radar measurements, coupling a bulk adaptive-habit model of vapor growth to a polarimetric radar forward model. Bayesian inference on key microphysical model parameters is then used, via a Markov chain Monte Carlo sampler, to estimate the probability distribution of the model parameters. The statistical formalism of this method allows for robust estimates of the optimal parameter values, along with (non-Gaussian) estimates of their uncertainty. To demonstrate this framework, observations from Department of Energy radars in the Arctic during a case of pristine ice precipitation are used to constrain vapor deposition parameters in the adaptive habit model. The resulting parameter probability distributions provide physically plausible changes in ice particle density and aspect ratio during growth. A lack of direct constraint on the number concentration produces a range of possible mean particle sizes, with the mean size inversely correlated to number concentration. Consistency is found between the estimated inherent growth ratio and independent laboratory measurements, increasing confidence in the parameter PDFs and demonstrating the effectiveness of the radar measurements in constraining the parameters. Furthermore, the combined Doppler and polarimetric observations produce the highest-confidence estimates of the parameter PDFs, with the Doppler measurements providing a stronger constraint for this case.

54 ENVIRONMENTAL SCIENCES↗

Optimal experimental design: Formulations and computations

Questions of ‘how best to acquire data’ are essential to modelling and prediction in the natural and social sciences, engineering applications, and beyond. Optimal experimental design (OED) formalizes these questions and creates computational methods to answer them. This article presents a systematic survey of modern OED, from its foundations in classical design theory to current research involving OED for complex models. We begin by reviewing criteria used to formulate an OED problem and thus to encode the goal of performing an experiment. We emphasize the flexibility of the Bayesian and decision-theoretic approach, which encompasses information-based criteria that are well-suited to nonlinear and non-Gaussian statistical models. We then discuss methods for estimating or bounding the values of these design criteria; this endeavour can be quite challenging due to strong nonlinearities, high parameter dimension, large per-sample costs, or settings where the model is implicit. A complementary set of computational issues involves optimization methods used to find a design; we discuss such methods in the discrete (combinatorial) setting of observation selection and in settings where an exact design can be continuously parametrized. Finally we present emerging methods for sequential OED that build non-myopic design policies, rather than explicit designs; these methods naturally adapt to the outcomes of past experiments in proposing new experiments, while seeking coordination among all experiments to be performed. Throughout, we highlight important open questions and challenges.

97 MATHEMATICS AND COMPUTING↗

The impact of response measurement error on the analysis of designed experiments

This study considers the analysis of designed experiments when there is measurement error in the true response or so-called response measurement error. We consider both additive and multiplicative response measurement errors. Through a simulation study, we investigate the impact of ignoring the response measurement error in the analysis, that is, by using a standard analysis based on t-tests. In addition, we examine the role of repeat measurements in improving the quality of estimation and prediction in the presence of response measurement error. We also study a Bayesian approach that accounts for the response measurement error directly through the specification of the model, and allows including additional information about variability in the analysis. We consider the impact on power, prediction, and optimization. Copyright © 2015 John Wiley & Sons, Ltd.

97 MATHEMATICS AND COMPUTING↗

Meas. K+ -Ar Total Inelastic Cross Section at ProtoDUNE-SP

ProtoDUNE-SP is a single-phase liquid argon time projection chamber that took hadron test beam data in 2018. The test beam included positively charged kaons with test beam momenta of 6 GeV/c and 7 GeV/c, providing a sample to study kaons to benefit future DUNE proton decay and neutrino interaction studies with kaons in the final state. The total inelastic cross section of a positively charged kaon was measured at these test beam settings using the LArIAT thin-slice method of dividing the wires of the time projection chamber into target slices for calculating the cross section, which leverages the monolithic quality of liquid argon detectors. A Bayesian-like unfolding method using RooUnfold was applied to both the incident and interacting slice distributions to measure the cross sections at both test beam momenta. The talk will discuss the method of unfolding, optimization studies for unfolding, and applying systematic uncertainties using a LArIAT-style hadronic cross section using unfolding to extract a kaon total inelastic cross section.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Multimodal parameter spaces of a complex multi-channel neuron model

One of the most common types of models that helps us to understand neuron behavior is based on the Hodgkin–Huxley ion channel formulation (HH model). A major challenge with inferring parameters in HH models is non-uniqueness: many different sets of ion channel parameter values produce similar outputs for the same input stimulus. Such phenomena result in an objective function that exhibits multiple modes (i.e., multiple local minima). This non-uniqueness of local optimality poses challenges for parameter estimation with many algorithmic optimization techniques. HH models additionally have severe non-linearities resulting in further challenges for inferring parameters in an algorithmic fashion. To address these challenges with a tractable method in high-dimensional parameter spaces, we propose using a particular Markov chain Monte Carlo (MCMC) algorithm, which has the advantage of inferring parameters in a Bayesian framework. The Bayesian approach is designed to be suitable for multimodal solutions to inverse problems. We introduce and demonstrate the method using a three-channel HH model. We then focus on the inference of nine parameters in an eight-channel HH model, which we analyze in detail. We explore how the MCMC algorithm can uncover complex relationships between inferred parameters using five injected current levels. The MCMC method provides as a result a nine-dimensional posterior distribution, which we analyze visually with solution maps or landscapes of the possible parameter sets. The visualized solution maps show new complex structures of the multimodal posteriors, and they allow for selection of locally and globally optimal value sets, and they visually expose parameter sensitivities and regions of higher model robustness. We envision these solution maps as enabling experimentalists to improve the design of future experiments, increase scientific productivity and improve on model structure and ideation when the MCMC algorithm is applied to experimental data.

97 MATHEMATICS AND COMPUTING↗

Bayesian estimation of Karhunen–Loève expansions; A random subspace approach

One of the most widely-used statistical procedures for dimensionality reduction of high dimensional random fields is Principal Component Analysis (PCA), which is based on the Karhunen-Lo eve expansion (KLE) of a stochastic process with finite variance. The KLE is analogous to a Fourier series expansion for a random process, where the goal is to find an orthogonal transformation for the data such that the projection of the data onto this orthogonal subspace is optimal in the L 2 sense, i.e, which minimizes the mean square error. In practice, this orthogonal transformation is determined by performing an SVD (Singular Value Decomposition) on the sample covariance matrix or on the data matrix itself. Sampling error is typically ignored when quantifying the principal components, or, equivalently, basis functions of the KLE. Furthermore, it is exacerbated when the sample size is much smaller than the dimension of the random field. In this paper, we introduce a Bayesian KLE procedure, allowing one to obtain a probabilistic model on the principal components, which can account for inaccuracies due to limited sample size. The probabilistic model is built via Bayesian inference, from which the posterior becomes the matrix Bingham density over the space of orthonormal matrices. We use a modified Gibbs sampling procedure to sample on this space and then build a probabilistic Karhunen-Lo eve expansions over random subspaces to obtain a set of low-dimensional surrogates of the stochastic process. We illustrate this probabilistic procedure with a finite dimensional stochastic process inspired by Brownian motion.

97 MATHEMATICS AND COMPUTING↗

An efficient Bayesian data-worth analysis using a multilevel Monte Carlo method

Improving the understanding of subsurface systems and thus reducing prediction uncertainty requires collection of data. As the collection of subsurface data is costly, it is important that the data collection scheme is cost-effective. Design of a cost-effective data collection scheme, i.e., data-worth analysis, requires quantifying model parameter, prediction, and both current and potential data uncertainties. Assessment of these uncertainties in large-scale stochastic subsurface hydrological model simulations using standard Monte Carlo (MC) sampling or surrogate modeling is extremely computationally intensive, sometimes even infeasible. In this paper, we propose an efficient Bayesian data-worth analysis using a multilevel Monte Carlo (MLMC) method. Compared to the standard MC that requires a significantly large number of high-fidelity model executions to achieve a prescribed accuracy in estimating expectations, the MLMC can substantially reduce computational costs using multifidelity approximations. Since the Bayesian data-worth analysis involves a great deal of expectation estimation, the cost saving of the MLMC in the assessment can be outstanding. While the proposed MLMC-based data-worth analysis is broadly applicable, we use it for a highly heterogeneous two-phase subsurface flow simulation to select an optimal candidate data set that gives the largest uncertainty reduction in predicting mass flow rates at four production wells. Finally, the choices made by the MLMC estimation are validated by the actual measurements of the potential data, and consistent with the standard MC estimation. But compared to the standard MC, the MLMC greatly reduces the computational costs.

97 MATHEMATICS AND COMPUTING↗

Formal Experimentation and Analysis of Handheld RFID Readers as a Tool for Nuclear Material Accounting

Commercial, off-the-shelf Radio Frequency Identification (RFID) systems have been successfully deployed for inventory tracking in numerous industries, but their viability in the tracking of complex environments containing nuclear material is less understood. Of primary interest in this setting is a RFID tracking system for nuclear material accounting which can reliably identify as many tags as possible with high accuracy, while also reducing an operator’s exposure to radiation. In this work, we develop a formal statistical approach to identify relevant handheld RFID reader settings which optimize tagging performance. To achieve this goal, we design a full factorial split-plot experiment for a static shelf configuration scene with 50 randomly placed nuclear material containers affixed with RFID tags. We use Bayesian inference to fit a second-order response surface model which expresses the probability of a successful match for each container as a function of the experimental factors. Such effects are allowed to vary across individual containers and the containers’ population in its entirety to estimate overall effects. Uncertainties of estimates and predictions are quantified via their corresponding posterior distributions. Following extensive model checking and validation, the fitted model is used to identify experimental factors which maximize matching probabilities at both the container-level and for the full shelf configuration scene. We also analyze sensitivity of performance to relevant factors.

25 ENERGY STORAGE↗

Revisiting thermodynamics and kinetic diffusivities of uranium–niobium with Bayesian uncertainty analysis

In this paper, thermodynamic and kinetic diffusivities of uranium–niobium (U–Nb) are re-assessed by means of the CALPHAD (CALculation of PHAse Diagram) methodology. In order to improve the consistency and reliability of the assessments, first-principles calculations are coupled with CALPHAD. In particular, heats of formation of γ -U–Nb are estimated and verified using various density-functional theory (DFT) approaches. These thermochemistry data are then used as constraints to guide the thermodynamic optimization process in such a way that the mutual-consistency between first-principles calculations and CALPHAD assessment is satisfactory. In addition, long-term aging experiments are conducted in order to generate new phase equilibria data at the γ 2 /α+γ 2 boundary. These data are meant to verify the thermodynamic model. Assessment results are generally in good agreement with experiments and previous calculations, without showing the artifacts that were observed in previous modeling. The mutual-consistent thermodynamic description is then used to evaluate atomic mobility and diffusivity of γ-U–Nb. Finally, Bayesian analysis is conducted to evaluate the uncertainty of the thermodynamic model and its impact on the system's phase stability.

36 MATERIALS SCIENCE↗

Bayesian and LASSO Regressions for Comparative Permeability Modeling of Sandstone Reservoirs

In this research, Bayesian model averaging (BMA) and least absolute shrinkage and selection operator regression (LASSO) algorithms were adopted for the core permeability modeling as a function of well log and core measurements. More specifically, the core permeability (dependent factor) was modeled given the well log and core data (independent variables) and then predicted in non-cored intervals of a well in a sandstone formation. The BMA is a stochastic linear modeling and a Bayesian parameter selection. Among 50 linear models generated as a function of the independent variables and depicted in Occam’s window, the best model of the highest posterior probability is determined. The best model has the optimal subset parameters that most influences the response factor. The core permeability modeling was again conducted using the LASSO algorithm, which adopts a completely different way of modeling and subset selection by using the penalized least-squared equation. Both BMA and LASSO resulted in a very accurate prediction of the core permeability by achieving perfect measured and calculated permeability matching. Matches for the two algorithms were quantified by computing the adjusted R-squared and root-mean-squared prediction error. In addition, results of the BMA and LASSO were compared to the conventional multiple linear regression (LM). The LASSO algorithm led to an accurate matching similar to LM, but slightly better than BMA. Therefore, both BMA and LASSO represent integrated procedures for accurate permeability estimation similar to conventional regression analysis.

58 GEOSCIENCES↗

mBEEF-vdW: Robust fitting of error estimation density functionals

Here, we propose a general-purpose semilocal/nonlocal exchange-correlation functional approximation, named mBEEF-vdW. The exchange is a meta generalized gradient approximation, and the correlation is a semilocal and nonlocal mixture, with the Rutgers-Chalmers approximation for van der Waals (vdW) forces. The functional is fitted within the Bayesian error estimation functional (BEEF) framework. We improve the previously used fitting procedures by introducing a robust MM-estimator based loss function, reducing the sensitivity to outliers in the datasets. To more reliably determine the optimal model complexity, we furthermore introduce a generalization of the bootstrap 0.632 estimator with hierarchical bootstrap sampling and geometric mean estimator over the training datasets. Using this estimator, we show that the robust loss function leads to a 10% improvement in the estimated prediction error over the previously used least-squares loss function. The mBEEF-vdW functional is benchmarked against popular density functional approximations over a wide range of datasets relevant for heterogeneous catalysis, including datasets that were not used for its training. Overall, we find that mBEEF-vdW has a higher general accuracy than competing popular functionals, and it is one of the best performing functionals on chemisorption systems, surface energies, lattice constants, and dispersion. We also show the potential-energy curve of graphene on the nickel(111) surface, where mBEEF-vdW matches the experimental binding length. mBEEF-vdW is currently available in gpaw and other density functional theory codes through Libxc, version 3.0.0.

36 MATERIALS SCIENCE↗

Trustworthiness modeling and evaluation for a nearly autonomous management and control system

The Nearly Autonomous Management and Control (NAMAC) system supports the advanced reactor operation by recommending control actions to operators based on real-time measurements and digital twins (DTs) learning from the knowledge base. To enable the safe and reliable use of autonomous technologies, NAMAC and its recommendations should be trustworthy to operators and regulators at both the design and operation stages. This study proposes a NAMAC trustworthiness modeling and evaluation framework supported by trustworthiness ontologies and evidence-based approaches. The development-time and run-time ontologies are separately constructed and then converted to Bayesian networks to quantitatively evaluate the NAMAC trustworthiness. This evaluation is demonstrated by collecting and characterizing evidence from NAMAC practices, such as the development and assessment of the NAMAC system, data coverage assessment, and the training and optimizations of neural-network-based DTs. Our proposed approach can aggregate various trustworthiness attributes of complex artificial-intelligence-supported systems for safety-critical applications. It also considers the interaction between different DTs and extends beyond the trustworthiness evaluation of a single DT. In conclusion, the evidence-based method enhances the transparency of the trustworthiness modeling and evaluation processes and helps identify uncertainties and subjectivity involved in the processes.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Gaussian processes with built-in dimensionality reduction: Applications to high-dimensional uncertainty propagation

Uncertainty quantification (UQ) tasks, such as model calibration, uncertainty propagation, and optimization under uncertainty, typically require several thousand evaluations of the underlying computer codes. To cope with the cost of simulations, one replaces the real response surface with a cheap surrogate based, e.g., on polynomial chaos expansions, neural networks, support vector machines, or Gaussian processes (GP). However, the number of simulations required to learn a generic multivariate response grows exponentially as the input dimension increases. This curse of dimensionality can only be addressed, if the response exhibits some special structure that can be discovered and exploited. A wide range of physical responses exhibit a special structure known as an active subspace (AS). An AS is a linear manifold of the stochastic space characterized by maximal response variation. The idea is that one should first identify this low dimensional manifold, project the high-dimensional input onto it, and then link the projection to the output. If the dimensionality of the AS is low enough, then learning the link function is a much easier problem than the original problem of learning a high-dimensional function. The classic approach to discovering the AS requires gradient information, a fact that severely limits its applicability. Furthermore, and partly because of its reliance to gradients, it is not able to handle noisy observations. The latter is an essential trait if one wants to be able to propagate uncertainty through stochastic simulators, e.g., through molecular dynamics codes. In this work, we develop a probabilistic version of AS which is gradient-free and robust to observational noise. Our approach relies on a novel Gaussian process regression with built-in dimensionality reduction. In particular, the AS is represented as an orthogonal projection matrix that serves as yet another covariance function hyper-parameter to be estimated from the data. To train the model, we design a two-step maximum likelihood optimization procedure that ensures the orthogonality of the projection matrix by exploiting recent results on the Stiefel manifold, i.e., the manifold of matrices with orthogonal columns. The additional benefit of our probabilistic formulation, is that it allows us to select the dimensionality of the AS via the Bayesian information criterion. We validate our approach by showing that it can discover the right AS in synthetic examples without gradient information using both noiseless and noisy observations. We demonstrate that our method is able to discover the same AS as the classical approach in a challenging one-hundred-dimensional problem involving an elliptic stochastic partial differential equation with random conductivity. Finally, we use our approach to study the effect of geometric and material uncertainties in the propagation of solitary waves in a one dimensional granular system.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Viscous flow properties and hydrodynamic diameter of phenothiazine-based redox-active molecules in different supporting salt environments

We report viscous flow properties of a redox-active organic molecule, N-(2-(2-methoxyethoxy)ethyl)phenothiazine (MEEPT), a candidate for non-aqueous redox flow batteries, and two of its radical cation salts. A microfluidic viscometer enabled the use of small sample volumes in determining viscosity as a function of shear rate and concentration in the non-aqueous solvent, acetonitrile, both with and without supporting salts. All solutions tested show Newtonian behavior over shear rates of up to 30 000 s−1, which was rationalized by scaling arguments for the diffusion-based relaxation time of a single MEEPT molecule without aggregation. Neat MEEPT is flowable but with a large viscosity (412 mPa⋅s at room temperature), which is ∼1000 times larger than that of acetonitrile. MEEPT solutions in acetonitrile have low viscosities; at concentrations up to 0.5 M, the viscosity increases by less than a factor of two. From concentration-dependent viscosity measurements, molecular information was inferred from intrinsic viscosity (hydrodynamic diameter) and the Huggins coefficient (interactions). Model fit credibility was assessed using the Bayesian Information Criterion. It is found that the MEEPT and its charged cations are “flowable” and do not flocculate at concentrations up to 0.5 M. MEEPT has a hydrodynamic diameter of around 8.5 Å, which is almost insensitive to supporting salt and state of charge. This size is comparable to molecular dimensions of single molecules obtained from optimized structures using density functional theory calculations. The results suggest that MEEPT is a promising candidate for redox flow batteries in terms of its viscous flow properties.

Wang, Yilin (ORCID:0000000316102995)↗